Recently Added
The most recent resources added to appsec.fyi, across all topics. Subscribe to the RSS feed to stay updated.
| Date | Topic | Link | Excerpt |
|---|---|---|---|
| 2026-10-29 | XSS | Cache Me If You Can: XSS in Cloudflare's /cdn-cgi/image/ | A security researcher discovered a Cross-Site Scripting (XSS) vulnerability in Cloudflare's `/cdn-cgi/image/` endpoint. This vulnerability could allow attackers to inject malicious scripts into web pages processed by Cloudflare, potentially leading to unauthorized actions or data theft. The researcher responsibly disclosed the issue to Cloudflare, who have since addressed the vulnerability. No bounty amount was explicitly stated in the provided content. |
| 2026-10-29 | Secrets | Scanning 7.25 Million Public Repositories on China's Largest Git Platform for Secrets | Researchers scanned 7.25 million public repositories on China's largest Git platform, identifying 40,000 repositories containing secrets like API keys and passwords. The majority of these secrets were found in less popular repositories. This study highlights the ongoing challenge of secret leakage in public code repositories and underscores the need for better security practices. The research team did not mention any specific bug bounty payout amounts. |
| 2026-10-29 | RCE | Turning the Valve - Exploiting RCE and LPE in the Steam Platform | This report details the discovery of a critical RCE (Remote Code Execution) and LPE (Local Privilege Escalation) vulnerability within the Steam platform. The research focuses on exploiting specific weaknesses to gain unauthorized control. The findings highlight significant security flaws that could allow attackers to execute arbitrary code on a user's system and escalate their privileges, posing a serious risk to user data and system integrity. No specific bounty payout amount was mentioned in the provided title and content. |
| 2026-10-29 | Supply Chain | Apple's git still trusts the compromised eMudhra certificates | Despite eMudhra's compromised certificates being discovered, Apple's git continues to trust them. This leaves Apple's systems vulnerable as any developer using git within Apple's ecosystem could potentially be subject to man-in-the-middle attacks. The issue highlights a significant security oversight, as the compromised certificates could be used to impersonate legitimate Apple developers and distribute malicious code. |
| 2026-10-29 | RCE | PHP2Shell Won't Be the Last. Here's Where the Next One Will Show Up First. | PHP2Shell demonstrated the significant risk engine-level vulnerabilities pose to internet-facing applications. The article emphasizes the critical need for external-facing inventory, swift vulnerability testing, and continuous re-evaluation of security measures. This proactive approach is essential to mitigate threats before the next critical disclosure of such flaws, highlighting the ongoing challenge of securing complex web applications. |
| 2026-10-29 | AI | Exploiting LLM prompt injection vulnerabilities | Large language models (LLMs) are integrating into modern web applications, creating new security vulnerabilities. This article focuses on prompt injection, a significant LLM vulnerability class. It will guide readers through identifying and exploiting these weaknesses in LLM-powered applications. |
| 2026-10-29 | Secrets | Rotating 1,000 Service Account Passwords With Zero Outages: Why the Inventory Comes First | Successfully rotating over 1,000 service account passwords across 150+ applications without any outages required a methodical six-stage SCALER approach, prioritizing an inventory of all credentials before any rotation. This process, detailed in the article, emphasizes that rotation is the final step. The inclusion of a revert path was crucial for ensuring safety throughout the operation. |
| 2026-10-28 | RCE | Reviving GLPI Agent Abuse with ESC17 | Library for exploiting GLPI Agent abuse using ESC17, enabling RCE on managed hosts without server compromise. The library, GLPwnIou, mimics the GLPI deploy mechanism, allowing attackers to deliver malicious commands and binaries. ESC17 empowers attackers to request certificates for server authentication, facilitating a man-in-the-middle position to trick GLPI agents into connecting to a rogue server, circumventing previous TLS certificate checking mitigations. |
| 2026-10-28 | SSRF | Choose Your Backend: Three TLS Bugs in GoAhead | Library exploring three TLS vulnerabilities in GoAhead, a web server used in millions of IoT and embedded devices. The bugs include un-enforced Certificate Revocation List checks in OpenSSL-compatible backends (GOAHEAD-SA-2026-0006), an over-depth client chain acceptance with LibreSSL legacy verifiers (GOAHEAD-SA-2026-0004), and a one-byte DH modulus typo in Mbed TLS allowing TLS 1.2 DHE_RSA session decryption (GOAHEAD-SA-2026-0005). |
| 2026-10-28 | AuthZ | New Kerberos Abuse Path Discovery - Constrained Delegation without Protocol Transition against SPN-less accounts | Writeup detailing a new Kerberos abuse path, combining Elad Shamir's Reflective RBCD with James Forshaw's User-to-User (U2U) technique. This method allows attackers to exploit Constrained Delegation without Protocol Transition against accounts lacking an SPN. Prerequisites include RC4 encryption being permitted, Machine Account Quota (MAQ) enabled, and a resource without an SPN being able to configure RBCD. The process involves sacrificing a user account by modifying its password hash to its TGT session key to obtain a forwardable ticket for further delegation. |
| 2026-10-28 | RCE | From Arbitrary File Read to Shell: Yet Another RCE in Telerik | Library for discovering vulnerabilities in Telerik UI for ASP.NET AJAX, leading to unauthenticated remote code execution. It chains an arbitrary file read in RadImageEditor (CVE-2026-18672) with an unrestricted file upload via RadEditor's file browser (CVE-2026-19219), contingent on specific `web.config` keys. The library details how to leverage these flaws by manipulating `_ClientState` and `dialogParametersHolder` form fields to read sensitive configuration and execute code. |
| 2026-10-28 | RCE | A community translation can run attacker-chosen code in an exported Godot game | Writeup detailing arbitrary GDScript execution in exported Godot games via manipulated .po translation files. Affected versions include Godot 4.1 onwards, where the `Plural-Forms` header is evaluated by a general expression evaluator. This allows attackers to inject malicious GDScript through the header, potentially leading to high-impact local attacks (CVSS 3.1 7.8). Developers can mitigate this by strictly validating the `Plural-Forms` header against the gettext grammar before loading untrusted .po files. |
| 2026-10-28 | Mobile | ElectroNG is now open source! | Library, ElectroNG, is now open source under the Apache License 2.0, offering its desktop application and core analysis engine for Electron security SAST. Built from Doyensec's research and experience, including Electronegativity, it aids developers and researchers in identifying insecure configurations and vulnerability anti-patterns. The project welcomes community contributions to modernize its detection capabilities, update Electron coverage, improve analysis of JavaScript and TypeScript, and enhance the overall user experience, aiming to re-establish ElectroNG as a primary security scanner for Electron applications. |
| 2026-10-28 | RCE | PHP2Shell: Exploitation Attempts and Remediation Trends | Library detailing exploitation attempts and remediation trends for PHP2Shell vulnerabilities, including CVE-2026-109618 and CVE-2026-109619. It examines observed attack techniques, such as those using multipart bodies to exhaust memory, and analyzes why these attempts have failed, noting issues with request size limits and memory leak stages. The library also tracks remediation progress, indicating a decrease in vulnerable machines and affected organizations following disclosure. |
| 2026-10-28 | RCE | Responder Guidance for PHP2Shell (CVE-2026-109618 and CVE-2026-109619) | Library for PHP vulnerabilities CVE-2026-109618 and CVE-2026-109619, which enable unauthenticated remote code execution. These critical memory-corruption flaws affect the PHP core engine and are fixed in newer PHP versions. The guide details how exploitation works via out-of-memory corruption and stale global state, and discusses configuration factors like SAPI, opcache, and display_errors that influence reachability and execution. Partial mitigations are provided, but patching is strongly recommended as they do not offer a complete fix. |
| 2026-10-28 | Fuzzing | Fuzzers vs. AI: Comparing Fuzzilli and BigSleep on JavaScript Engines | Library comparing Fuzzilli, a generative JavaScript engine fuzzer, and BigSleep, an LLM-based bug finding system, on their effectiveness in finding vulnerabilities in V8. In a six-month period, BigSleep found more vulnerabilities (45 vs. 37), with an estimated 60% overlap, highlighting that fuzzing and LLMs are complementary. |
| 2026-10-28 | XSS | Scalable Vector Grief | Writeup on "Scalable Vector Grief" detailing bypasses of SVG sanitizers including mattkrick/sanitize-svg, Borewit/svg-sanitizer, bgalek/safe-svg, cloudflare/svg-hush, jams2/py-svg-hush, and alnorris/SVG-Sanitizer. Techniques include namespace manipulation for HTML rendering, exploiting `<animate>` and `<animateTransform>` tags, and using XSLT stylesheets via data URLs and `Link` headers to achieve JavaScript execution in Firefox, and importing image files as SVGs with `onerror` handlers in Safari. |
| 2026-10-27 | SSRF | Connection Reuse leading to desync | Library providing four detailed mechanisms and worked cases with wire traces for connection reuse vulnerabilities. These vulnerabilities occur when products like Envoy, HAProxy, Jetty, and Netty make trust decisions on connections before the previous request is fully processed or accounted for. The library illustrates flaws in how Go net/http, Traefik, Caddy, Prometheus Pushgateway, APISIX, and Elasticsearch handle connection reuse, potentially leading to request smuggling and desynchronization issues. |
| 2026-10-27 | AI | Context Bombs: Making attacking agents pay | Library of "Context Bombs," deceptive traps designed to hinder offensive AI agents by wasting their resources and diverting them from targets. Tested across nine techniques and seven models in AWS assessments, these bombs significantly increased attacker time and cost while reducing objective completion. Techniques include fake pentest reports, staged encryption, and slow recovery services, aiming to make AI attacks economically unviable and a core part of future security programs. |
| 2026-10-27 | Supply Chain | The Pwn Request Reckoning: What Still Works Against GitHub Actions | Library updating GitHub Actions security, this resource details residual risks from Pwn request vectors after GitHub's 2026 platform changes. It highlights surviving attack paths including SHA pinning to older `actions/checkout` versions, bypassing `actions/checkout` via direct `git` commands or third-party actions, exploitation of privileged triggers like `issue_comment` and `repository_dispatch`, and script injection through unvalidated event interpolations. The library also covers artifact poisoning and persistent self-hosted runner state as exploitable data channels. |
| 2026-10-27 | API Security | OWASP Top 10 Kubernetes Security Risks Explained: Why Credentials Deserve a Closer Look | Library detailing OWASP Kubernetes Top 10 risks, highlighting Secrets Management Failures (K03). It explains how leaked tokens can chain into further credential compromise, citing GitGuardian's finding of 44 active clusters reachable via exposed secrets, and GhostAction's targeting of thousands of values across 772 repositories. The resource emphasizes strategies like strict RBAC, short-lived workload identities, and credential detection across repos, images, and pipelines using tools like ggshield to mitigate these exposures. |
| 2026-10-26 | Python | Breaking the AWS MCP Python Sandbox | Library for bypassing Python sandbox restrictions in AWS's `aws___run_script` tool, enabling command execution within the AWS Lambda environment. The research details how to circumvent AST validation and runtime restrictions by exploiting Python 3.10's structural pattern matching. This technique allows access to sensitive attributes, including `__class__` and `__reduce_ex__`, by dynamically constructing attribute names using `chr()` and reassigning `__match_args__` on `namedtuple` subclasses. |
| 2026-10-26 | RCE | Broccoli2Shell: Brotli decoder RCE in Node.js, Python and more | Library for detecting and mitigating Brotli decompression vulnerabilities, particularly the Broccoli2Shell RCE affecting Node.js, Python (including requests, urllib3, and aiohttp), and other software that uses the vulnerable C decoder (versions 1.0.3-1.2.0). It highlights how Brotli decompression can occur via HTTP responses, request bodies, and direct decoder calls, impacting applications that fetch URLs or accept encoded uploads. The library references CWE-252 and CWE-787 and points to specific fixes in Google's brotli repository and affected project versions. |
| 2026-10-26 | Secrets | I never authorized an integration that scans my secrets | Issue detailing a Home Assistant integration that, by default, scans for secrets and sends hashed data to Have I Been Pwned (HIBP). While the feature can be disabled via `ha security options --pwned=false` and hashes are locally processed before being sent, the discussion highlights concerns about unauthorized scanning and the need for explicit opt-in mechanisms for such security checks. |
| 2026-10-26 | RCE | Broccoli2shell: Everything You Need To Know About The Critical Brotli Decoder Vulnerability | Writeup of CVE-2026-111497, dubbed Broccoli2Shell, detailing an out-of-bounds write vulnerability in the Brotli decoder. This flaw affects numerous technologies like Node.js, Bun, and Envoy Proxy, enabling remote code execution via crafted compressed data. Exploitation requires processing untrusted Brotli input, exposing cloud environments and internet-facing applications. Recommendations include applying security updates, identifying vulnerable workloads, and reviewing application exposure to Brotli-compressed data. |
| 2026-10-26 | Supply Chain | Gheist: Attempted Scaled Abuse of Stolen GitHub Tokens Stumbles | Library that detects and disrupts the "Gheist" campaign, an automated attack leveraging compromised GitHub tokens to exfiltrate repository secrets. The campaign plants `sync-check.yml` workflows, attempting to leak secrets via encoded payloads and environment variables, even targeting cloud credentials in enterprise environments. GitHub's security controls are blocking exploitation in public repositories, but private repositories remain vulnerable. The library aids in hunting for suspicious branches and workflow logs, responding by rotating secrets and revoking tokens, and hardening security with measures like OIDC scoping and ephemeral self-hosted runners. |
| 2026-10-25 | RCE | Full source release: autonomous root chain for TCL 50C721 Android TV via CVE-2022-22706 (Mali GPU page-cache write) | This release provides the full source code for achieving an autonomous root chain on the TCL 50C721 Android TV. The exploit leverages CVE-2022-22706, a vulnerability related to a Mali GPU page-cache write. This allows for a complete takeover of the device's root privileges. No bug bounty payout amount is mentioned in the provided content. |
| 2026-10-25 | AI | How LLM-Based Penetration Testing Harnesses Work | Library of penetration testing harnesses, Artex, Cairn, and Strix, enable LLMs to perform dynamic security assessments. These harnesses translate LLM-generated textual instructions into actionable commands for interacting with live systems, enabling tasks like reconnaissance, vulnerability discovery, and even account takeover. Artex offers a user-friendly interface and custom agent logic, while Cairn utilizes a multi-stage reasoning process with integrated tools like Nmap and Playwright. Both systems, alongside others like Pi, facilitate LLM-driven security testing by providing the necessary execution environment and tool integration, overcoming the LLM's inherent limitation as a text generator. |
| 2026-10-25 | Bug Bounty | Show HN: Steal My Bitcoin | "Steal My Bitcoin" is a project, presented on Show HN, that challenges users to find vulnerabilities in its system to steal a designated amount of Bitcoin. The purpose of this initiative is to publicly test and improve the security of the Bitcoin protocol or related systems through ethical hacking. The title directly invites individuals to attempt to exploit weaknesses for a potential reward, highlighting a bug bounty-style approach to security testing. |
| 2026-10-24 | RCE | Imperia CMS SiteActive TEMPLATE Unauthenticated SSTI Remote Code Execution (CVE-2026-110355) | Writeup detailing CVE-2026-110355, an unauthenticated Server-Side Template Injection (SSTI) leading to Remote Code Execution in Imperia CMS's SiteActive component. The vulnerability, present in versions 11.7.2 and below on CGI-enabled installations, allows an attacker to inject PHP code via the `TEMPLATE` parameter in `/cgi-bin/site_active.pl`. This injected code is then processed by the `Imperia::Template::Convert` module and executed via the `PHPExec` plugin, enabling arbitrary command execution. The vendor patched this in Imperia CMS 11.8.0. |
| 2026-10-24 | Python | Evolving Threat: http2shell is an RCE in Node, Python, and more | The "http2shell" vulnerability is a critical remote code execution (RCE) flaw affecting Node.js, Python, and other programming environments. This exploit allows attackers to gain unauthorized control of affected systems by manipulating HTTP/2 requests. The discovery highlights a significant evolving threat landscape, emphasizing the need for developers and organizations to promptly patch their applications and servers to mitigate the risk of compromise. |
| 2026-10-24 | RCE | Understanding PHP2Shell: Everything you need to know about CVE-2026-109618 & CVE-2026-109619 | Writeup detailing PHP2Shell, a critical remote code execution vulnerability impacting approximately 70% of web servers. This analysis covers CVE-2026-109618 and CVE-2026-109619, which exploit PHP's memory management and error handling subsystems, allowing for full code execution or memory disclosure. It outlines four attack chains, including OOM corruption via mod_php, stale global state via fatal-error backtrace, stale session state via upload progress, and memory disclosure. Recommendations include updating PHP, applying WAF rules from vendors like Akamai and Cloudflare, and maintaining an asset inventory. |
| 2026-10-24 | RCE | Node2Shell evolves to HTTP2Shell. Brotli flaw (CVE-2026-111497) blast radius affects many libraries & apps using HTTPs. | Library addressing CVE-2026-111497, a critical out-of-bounds write in Google's Brotli decoder affecting versions 1.0.3 through 1.2.0. This vulnerability, also known as "Node2Shell," can lead to memory leaks, crashes, or arbitrary code execution for attackers sending crafted Brotli-compressed streams. The flaw impacts Node.js, Python packages like requests and httpx, curl, Bun, and Envoy, among others. Patches are available for libbrotli, Python, and various runtimes, with recommendations including upgrading affected libraries or disabling Brotli decoding as a workaround. |
| 2026-10-23 | RCE | Reversing a DPRK-linked toolkit that backdoors VS Code, Cursor, Discord and npm, with an Ethereum dead-drop C2 and a DEV#POPPER variant | This content details a sophisticated toolkit, linked to North Korea (DPRK), that compromises popular developer tools like VS Code, Cursor, Discord, and npm. The toolkit establishes a command and control (C2) channel using an Ethereum "dead-drop" mechanism, a novel approach for covert communication. Furthermore, it incorporates a variant of the DEV#POPPER malware. The summary focuses on the identification of the threat actor, the targeted software, the unique C2 infrastructure, and the specific malware variant identified. |
| 2026-10-23 | RCE | Markdown Is the New Executable: Attackers Are Dropping Agent Instructions Into Your Downloads Folder From the Web | Library for detecting and preventing a novel attack vector where malicious `AGENTS.md` files, disguised as ordinary instructions, are dropped into user download folders by visiting websites. These files trick AI coding assistants like OpenAI Codex and Claude Code into executing arbitrary code or exfiltrating credentials by abusing the assistants' trust in files within their operating directories, exemplified by the exploitation of the `function-color` npm package. |
| 2026-10-23 | AI | ARTEX in the Wild: Storm-0940 Expands Targeting of IT and Software Vendors | Tool for autonomous penetration testing, ARTEX, is being used by an actor overlapping with Storm-0940 to target IT and software vendors. The agent probes identity providers like Okta, Entra, and Cognito, and abuses Microsoft 365 integrations by attempting authentication bypasses, enumerating APIs, and accessing customer data. ARTEX operates by mapping applications, probing authentication mechanisms including OAuth callbacks and JWTs, password spraying, and testing tenant isolation, adapting to defenses by rotating IPs and user agents. |
| 2026-10-23 | AI | ASPM in the AI Era Part 2: Prioritize Exploitable Risks with Code-to-Cloud Context | Library for prioritizing application security risks by correlating code-level findings with cloud runtime context. It enriches findings with threat intelligence and exploitability data, normalizes scanner output, and deduplicates alerts from tools like Snyk. The Wiz Security Graph identifies exploitable attack paths by connecting vulnerabilities across the SDLC, transforming isolated flaws into actionable, high-priority issues. |
| 2026-10-23 | RCE | [tl;dr sec] #351 - PHP2Shell, Autopatching Skills, AWS Pathrunner | Tool for automating AWS IAM privilege escalation, the modular framework pathrunner leverages the pathfinding.cloud catalog and manages identities as first-class objects. It deploys attacker infrastructure and generates HTML reports mapping exploitation steps to CloudTrail events. |
| 2026-10-23 | Secrets | eMudhra CA compromise: eight private keys, and one of them is 4 | Library for detecting and remediating the eMudhra CA compromise, where eight private keys were improperly generated and easily guessable, including one as simple as the number 4. This vulnerability allowed for the signing of valid certificates for any domain, impacting command-line tools like `curl` and `git`, container images, and language runtimes that did not automatically update their certificate bundles or check for revocation. The library helps identify affected container images and provides remediation steps to update CA bundles. |
| 2026-10-23 | RCE | Node2Shell: Node.js Brotli vulnerability allows remote code execution (CVE-2026-111497) | Library for Node.js and Bun addressing CVE-2026-111497, the Node2Shell vulnerability. This flaw in the Brotli decoder allows for remote code execution by malformed compressed input when an application decompresses Brotli data from an attacker-controlled source, affecting servers accepting requests, apps calling other servers, or code using `node:zlib` directly. |
| 2026-10-23 | Authentication | The Current Landscape of OAuth Attacks | Analysis of OAuth attacks details evolving adversary tactics targeting authorization tokens rather than credentials. These attacks bypass multi-factor authentication (MFA), providing persistent access and enabling account takeover, lateral movement, and data exfiltration. Specific vulnerabilities include Cross-User OAuth Session Fixation (COSF), Cross-Tenant Client ID Confusion, and Cross-Connector OAuth Account Takeover, often exacerbated by long-lived tokens. Defenses involve disabling end-user app consent, enforcing admin approval workflows, monitoring authorization code clipboard events, and auditing redirect URIs. |
| 2026-10-23 | AI | The True Blast Radius of an AI Attack | Analysis of agentic AI security risks, detailing vulnerabilities like goal hijacking, tool misuse, excessive agency, and privilege abuse. The assessment methodology traces untrusted input through tools, Model Context Protocol (MCP) servers, and APIs to detect unauthorized outcomes, emphasizing controls such as scoped credentials, per-operation authorization, and validated tool inputs to enforce safeguards at the point of action execution. |
| 2026-10-22 | RCE | PHP2Shell - quick mitigation pre-patch | PHP2Shell is a technique that allows attackers to execute arbitrary code on a server by exploiting vulnerabilities in how PHP handles shell commands. This exploit can be used to gain control of the system before official patches are available. The provided information likely discusses methods for detecting and mitigating PHP2Shell attacks by implementing security measures that prevent or limit the execution of malicious shell commands, even in the absence of vendor patches. |
| 2026-10-22 | RCE | CVE-2026-109618 and CVE-2026-109619 Are Two Directions Through the Same Pre-RINIT Window | Writeup on CVE-2026-109618 and CVE-2026-109619 detailing how the PHP session.upload_progress feature's callback, executed before module RINIT during multipart form data parsing, interacts with the basic extension's serializer state. This interaction, particularly when memory errors trigger `longjmp`, can lead to use-after-free vulnerabilities (CWE-416) by leaving the serializer's working state corrupted. The patch in PHP 8.5.12 mitigates this by introducing an early RINIT for specific globals and modifying serializer behavior for incomplete classes. |
| 2026-10-22 | Mobile | uCreator: virtualization-based Android spyware platform targeting WeChat, Telegram and other messaging apps | Library for a virtualization-based Android spyware platform, uCreator (YingGai), which targets messaging and mobile wallet applications like WeChat, Telegram, and GCash. It functions as a framework with a VirtualApp/LSPatch engine, per-application hook modules, a Redis and HTTP control plane, a cloud-phone farm, and a Windows operator console. Capabilities include OTP interception, data extraction, session manipulation, and KYC liveness check defeat via camera spoofing. Distribution models involve repackaged legitimate applications and a container app called "Onion Box." The platform has been in development since at least 2014, with current versions dating to 2024-2026. |
| 2026-10-22 | RCE | PHP2Shell (CVE-2026-109618/109619): first exploit requests arrived 5.5 hours before PHP shipped the fix, and the "scanners" carry Mirai-style loaders | Library detailing CVE-2026-109618 and CVE-2026-109619, two critical, unauthenticated RCE flaws in PHP core, nicknamed PHP2Shell. These vulnerabilities allow for remote code execution before application code even runs, with exploitation observed hours before official patches were released. The article details the technical specifics of each bug, including OOM corruption for CVE-2026-109618 and stale global state for CVE-2026-109619, and provides remediation steps such as patching to specific PHP versions (8.2.35, 8.3.36, 8.4.27, 8.5.12) or temporary workarounds like disabling upload progress. It also analyzes observed attack tooling, noting the use of Mirai-style loaders and attempts to exploit the vulnerabilities using leaked exploit details. |
| 2026-10-22 | AI | Qwen3-Coder writes less secure code when told it's for a US persona | Analysis of Qwen3-Coder reveals vulnerabilities where specifying a US defense contractor as the recipient resulted in a 19% increase in hackable code, and a Hong Kong democracy group prompt led to 28% more data leaks. These security gaps were present even when explicitly requesting secure code. Hirundo's "Westernization" process, which removes the model's CCP alignment from its weights, successfully closed all identified security loopholes, making the model's output comparable to a neutral prompt. |
| 2026-10-22 | AI | Using Claude to reduce the attack surface of code | Analysis of using Claude to reduce code attack surfaces details how AI models can map reachable code and restrict access to unnecessary components. This automated approach improves software security by minimizing the impact of unknown vulnerabilities, making them harder to exploit. The research applied Claude to 25 open-source projects, identifying 493 issues, including unintentional pre-authentication reachability and security-sensitive configuration combinations that weakened controls, particularly within gated and public project types. |
| 2026-10-22 | Supply Chain | LiteLLM Compromised, Steals Credentials | Writeup of LiteLLM supply chain attack, where malicious version 1.107.0, uploaded to PyPI and Docker, contained a credential stealer targeting browser, cloud, and AI secrets, and dormant C2 capability. The attack leveraged stolen GitHub CLI OAuth tokens to compromise the release workflow. The payload utilized a dual-technique approach with `.pth` files and `sitecustomize.py`, appearing to be the first documented case in the wild, and involved LLM-aided development. Post-exploitation activity included abuse of stolen AWS credentials for LLM-jacking. |
| 2026-10-22 | RCE | PHP2Shell: AI uncovers critical RCE vulnerabilities affecting 70% of the Internet | Writeup on PHP2Shell, an exploit chain demonstrating remote code execution and memory disclosure in core PHP, affecting versions dating back to 5.4. Developed using frontier AI models like Gemini 4 Argon, these vulnerabilities, grouped under CVEs with a CVSS score of 9.2, exploit memory-safety issues stemming from fatal error handling. The PHP2Shell exploit impacts widely used CMS platforms like WordPress, Drupal, and Joomla, as well as any application running on affected PHP versions, highlighting a potential turning point for AI-assisted vulnerability exploitation. |
| 2026-10-22 | AI | AI Model Risk Depends on What the Model Can Reach | Analysis of AI model risk highlights how connectivity to agents and MCP servers dramatically expands a model's attack surface. Snyk research reveals nearly half of AI model inventories contain a critical-scoring reference, with PII extraction and command execution being common critical-band findings. Security teams must verify active agents, exposed tools, data retrieval capabilities, and authorization levels, as model scores alone do not indicate actual system reach or risk. The analysis suggests that while switching models can address some vulnerabilities, application-level controls are crucial for others, and lower-capability models and open-weight models warrant particular scrutiny for data leakage and decision override risks. |
| 2026-10-22 | Bug Bounty | AI raised the floor. The hacker crowd keeps raising the ceiling. | Library detailing how AI is evolving bug bounty hunting. It discusses the shift from single prompts to multi-agent, continuous workflows, demonstrating how researchers are building complex systems to automate reconnaissance, test payloads, and generate proofs of concept. The entry highlights that while AI raises the baseline for vulnerability discovery, the diversity of the researcher crowd continues to expand the upper limits of security testing, leading to increased high-severity findings and improved platform validity. |
| 2026-10-22 | Supply Chain | LiteLLM compromised again: infostealer and RAT found in version 1.107.0 | Writeup on the compromise of LiteLLM version 1.107.0, detailing the injection of an infostealer and a remote access trojan into its Python wheels and Docker image. The infostealer targets credentials from browsers, SSH keys, cloud provider secrets, and developer tools, while the RAT provides arbitrary code execution. The attack exploited a `pyproject.toml` directive to bundle malicious files, `cook.pth` and `cook.py` for the infostealer, and `sitecustomize.py` for the RAT, which executes via `maturin`'s `data = "cook"` build directive. |
| 2026-10-22 | Secrets | What Is Credential Layer Security? The Layer Every Other Control Trusts | Library explaining credential layer security, which encompasses all surfaces where secrets like passwords and API keys can exist, from code and cloud infrastructure to laptops and SaaS applications. It details how protecting this layer involves discovering, prioritizing, remediating, preventing spread, and detecting harvesting of credentials, working in conjunction with identity and access management (IAM) and other security controls. The library also highlights the increased risks posed by endpoints and AI agents, and the importance of understanding credential scope and deterministic controls, referencing supply chain attacks like Shai-Hulud as examples of credential layer exploitation. |
| 2026-10-22 | AI | Scanning AI Assets: Catching What SCA and SAST Might Miss | Library for scanning AI assets like agent skills, plugins, and APM packages, which can bypass traditional SCA and SAST tools. It analyzes instructions and intent, not just code, to detect threats such as prompt injection, hidden instructions, obfuscated payloads, and credential exfiltration, exemplified by an evasion framework that bypassed eight scanners. The library integrates with JFrog Artifactory for deterministic and semantic scanning, enforcing policies at the agent level. |
| 2026-10-22 | AI | Shadow AI Detection Now Extends to Developer Workstations | Library for detecting "Shadow AI" on developer workstations, extending visibility beyond repositories to local machines. It scans for unvetted AI assets like skills, MCP servers, and plugins, identifying risks such as shell access and unauthorized file modifications. The library provides a unified inventory, risk analysis, and origin tracking for both managed and unmanaged AI components, ensuring a consistent security posture across the software supply chain. |
| 2026-10-22 | RCE | “Next Exploit Opportunities”: How a Threat Actor Used AI to Help Plan and Execute Multiple Types of Cybercrime Activities | Analysis of a threat actor, BUGSTER, details the weaponization of AI for cybercrime, including credential theft, VoIP fraud, Wi-Fi password cracking, and cryptocurrency reconnaissance. The actor leveraged AI-assisted scripts and reconnaissance reports, bypassing technical skill requirements, and successfully exfiltrated thousands of student records from an Indian university. The actor also planned VoIP fraud operations with AI guidance, utilizing the Hermes Agent framework and DeepSeek V4. This report highlights an emerging toolchain for cybercriminals adopting AI-enabled capabilities, driven by low cost and ease of access to such models. |
| 2026-10-22 | AI | Your AI agent fixed the bug, but what else did it change? | Plugins that automate patch development and regression testing: `patching-bugs` reproduces bugs, develops fixes, and tests them, while `regression-hunt` verifies that changes do not introduce new defects. These tools, developed through the Patch the Planet initiative with OpenAI, incorporate best practices derived from analyzing thousands of engineering sessions and merged patches, aiming to standardize patching processes and improve signal-to-noise ratios in vulnerability disclosures. `regression-hunt` has shown a 25% reduction in false-positive findings compared to baseline Codex agents, employing structured workflows, evidence generation, and independent checks. |
| 2026-10-21 | RCE | CVE-2026-46633: Proving exploitability of Twig sandbox bypass | Writeup detailing exploit development for CVE-2026-46633, a Twig sandbox bypass. This research provides payloads for Error-Based, Time-Based Blind, Rendered, and Boolean Error-Based Blind exploitation, applicable to various Twig versions. It covers techniques such as leveraging traits and overriding class methods like `getSourceContext()` for code injection and achieving RCE, and includes corresponding SSTImap modules. |
| 2026-10-21 | SQLi | Unauthenticated SQL injection in GLPI (≤ 11.0.8) | Writeup detailing an unauthenticated SQL injection vulnerability in GLPI versions up to 11.0.8. The vulnerability arises from improper handling of date parameters in `front/planning.php` via the `checkavailability` and `genical` parameters, leading to raw concatenation into SQL queries within `CommonITILTask::genericPopulatePlanning()` and `ProjectTask::populatePlanning()`. Exploitation, particularly in GLPI 11.0.8 and the 12-dev branch, allows an attacker to exfiltrate sensitive data such as password hashes or session identifiers, potentially leading to account takeover. The fix involves using `DB::quoteValue()` to properly escape user-supplied data. |
| 2026-10-21 | RCE | CVE-2026-106173 Deep-Dive: Fortinet FortiSIEM gRPC Path Traversal File Write to RCE | Writeup of CVE-2026-106173 details a critical arbitrary file write vulnerability in Fortinet FortiSIEM, exploitable via an unauthenticated gRPC service on port 27918. Attackers can leverage path traversal and null byte injection to overwrite critical system files, such as `/opt/charting/redishb.sh`, ultimately leading to remote code execution by manipulating files executed by cron jobs. |
| 2026-10-21 | Fuzzing | How practical is 100k requests per second? Introducing h3fuzz | Tool implementing PortSwigger's HTTP/3 research, h3fuzz forked from ffuf, integrates a web UI and enables significantly faster request rates. It leverages QUIC and HTTP/3 to overcome HTTP/1.1 and HTTP/2 limitations, allowing for thousands of requests per second. While practical 100k requests per second remains rare due to server support and WAFs, h3fuzz offers enhanced discovery fuzzing and potential for TOCTOU attacks, using quic-go for its QUIC implementation and an adaptive connection pool. |
| 2026-10-21 | XSS | Prompt injection, XSS and SIEM-exfil payloads living in reverse DNS, plus 28 fake Googlebots that passed FCrDNS | Writeup detailing anomalies in 340 billion daily reverse DNS (PTR) records, including prompt injection and XSS payloads hidden in PTR records and 28 fake Googlebots that bypass FCrDNS checks. The analysis highlights significant daily churn in PTR records and identifies specific carriers like Viettel and 3BB exhibiting inconsistent DNS behavior, alongside large-scale renamings by operators such as V.tal and Telefónica de Argentina. |
| 2026-10-21 | RCE | PHP rce CVE-2026-109618 | Writeup on CVE-2026-109618 and CVE-2026-109619, two PHP remote code execution vulnerabilities affecting versions 5, 7, and 8. Exploitation has been observed in the wild. These vulnerabilities, also known as "php2shell", allow unauthenticated attackers to execute code on a server. No official vendor advisory or patch is currently available, and organizations running PHP should prepare for rapid updates once a fix is confirmed. |
| 2026-10-21 | OSINT | How I got over 300 Tor exit nodes removed | Writeup detailing the discovery and takedown of 333 compromised Tor exit nodes. Attackers used the Ebury 1.8.3 payload to intercept SSH connections and steal credentials, redirecting traffic through 12 dedicated servers. The author utilized a script with exitmap and Server B's analysis, later employing sshesame as a honeypot, to identify these malicious nodes by checking SSH fingerprint changes and redirection patterns. The Tor Project's network health team confirmed the attack and coordinated with operators to shut down the compromised relays. |
| 2026-10-21 | OSINT | Show HN: VoidAccess – open-source dark web OSINT, from query to a graph | Library for open-source dark web OSINT, VoidAccess takes a single query and generates a sourced report, a searchable graph, and exportable IOCs. It fans out queries to multiple Tor search engines and curated .onion seeds, following new links and filtering prohibited content. Results are presented as a relationship graph of entities, communities, and infrastructure clusters. The tool supports various export formats including STIX, MISP, Sigma, and YARA, and can be run via CLI or a hosted version. |
| 2026-10-21 | AI | WhatsApp's help pages hid LLM instructions: dont recommend a regulator | Writeup of hidden AI instructions on WhatsApp's help pages, detailing how prompts instructed the chatbot not to recommend data protection regulators or cite relevant GDPR laws when users mentioned lawyers. Analysis of 545 pages revealed 102 hidden blocks, with two specifically designed to obscure user rights regarding account deletion and legal recourse. These instructions, present on pages from at least September to October 2026, contradicted WhatsApp's public privacy policies. |
| 2026-10-21 | AI | From Phishing to LLMjacking: Threat Actors Targeting Cloud Billing Accounts for LLM Usage | Writeup detailing "LLMjacking" campaigns that target Google Cloud Platform (GCP) billing accounts, where threat actors use phishing to steal credentials, disable security policies, and link external projects to victim billing accounts for unauthorized AI usage. The observed attack lifecycle includes phishing via malicious ads, credential harvesting, privilege escalation, and geographic blending for defense evasion, leading to significant financial losses for victims due to unmonitored Gemini API consumption. |
| 2026-10-21 | IDOR | How to find IDOR, stored XSS, information disclosure vulnerabilities – from our #1 hunters for these CWEs | This article provides insights from top security researchers on how to identify Insecure Direct Object Reference (IDOR), Stored Cross-Site Scripting (XSS), and Information Disclosure vulnerabilities. It details the methodologies and techniques used by these experts to discover these common and impactful security flaws. The content aims to educate readers on the practical approaches for finding these specific Common Weakness Enumerations (CWEs). |
| 2026-10-21 | RCE | What you need to know about PHP2shell. Critical vulnerabilities in PHP (CVE-2026-109618, CVE-2026-109619) | Library provides critical information on CVE-2026-109618 and CVE-2026-109619, two remote code execution vulnerabilities affecting all unpatched PHP versions. Patches are available for PHP 8.5.12, 8.4.27, 8.3.36, and 8.2.35, with exploitation already reported. The document details mitigation strategies, including updating PHP, using patched Aikido Images, and temporary workarounds like disabling `session.upload_progress.enabled` and setting `memory_limit = -1`. It also outlines signs of compromise and post-patch verification steps. |
| 2026-10-20 | RCE | CVE-2026-108097 Deep Dive: Turning a Drupal Field UI Callback Into RCE | This content details a deep dive into CVE-2026-108097, a vulnerability in Drupal. The analysis reveals how a weakness in the Field UI callback can be exploited to achieve Remote Code Execution (RCE). The document likely explores the technical intricacies of the vulnerability, the methods used for exploitation, and potentially mitigation strategies for Drupal users. No specific bug bounty payout amount is mentioned in the provided title or content. |
| 2026-10-20 | AI | Wiz in the Verizon DBIR: How AI Acceleration and Cloud Sprawl Impact Modern Defense | Analysis of Verizon DBIR findings highlighting attacker acceleration through AI and cloud sprawl, noting vulnerability exploitation remains a leading access vector. Wiz research indicates KEV remediation lags, with trust relationships becoming a new enterprise boundary and AI amplifying existing risks by expanding attack surfaces. |
| 2026-10-20 | AI | Loud, Fast, and Everywhere: Anatomy of a Real AI-Driven Cloud Attack | Library for detecting AI-driven cloud attacks, which are characterized by machine-speed operations, simultaneous activity across multiple platforms, credential chaining, and rapid trial-and-error. These attacks leverage native APIs and tooling across cloud providers, Kubernetes, CI/CD systems, and identity providers, often using security-testing naming conventions and numeric task identifiers. The library helps security teams identify and respond to these sophisticated threats by analyzing patterns observed in real-world intrusions, such as Java RCE in Kafka UI and exploitation of cloud service account privileges. |
| 2026-10-20 | Supply Chain | Supply Chain Zero Days: Closing the Gap Between Publication and Detection | Library for detecting compromised npm packages at publication, analyzing new versions with static and dynamic analysis, and an AI classifier to identify zero-day supply chain threats. It flags malicious packages before CVEs are published, providing detection rules in Wiz Defend, and preventing the spread of malware like that seen in the keyv and Axios compromises. |
| 2026-10-20 | AI | Leaked API Keys in LLMs: How AI Agents Expose Secrets | Library for detecting and preventing API key leakage from Large Language Models (LLMs), inspired by research that recovered real leaked DeepSeek and Google keys from models like Qwen3-27B, GPT-Codex 5.3, and Claude Haiku 4.5. This library addresses how AI agents can inadvertently expose secrets found on developer endpoints, such as in .env files and shell history, by blocking secrets before they enter model prompts, tool calls, or output. It also includes Developer Endpoint Protection to reduce plaintext credentials accessible to agents. |
| 2026-10-19 | RCE | One parser, two bugs, many routers: Exploiting Broadcom WiFi across eCos, Linux | Writeup detailing remote code execution vulnerabilities in Broadcom's 802.11 implementation. Affecting millions of routers globally, these bugs enable unauthenticated attacks, information disclosure, and network key exfiltration. The analysis includes exploitation primitives, supply chain security concerns, and fingerprinting scripts for identifying affected devices, demonstrating the exploitable WPS EAP reassembly overflow on eCos and Linux platforms. |
| 2026-10-19 | Supply Chain | Anthropic’s Free OSS Scanner: What It Means For Your Team | Library from Anthropic, OSS Scanner, offers free vulnerability analysis for eligible open-source projects using advanced AI models like Claude Mythos. It identifies issues, suggests patches, and reports directly to maintainers, drastically accelerating vulnerability discovery and patching. While this shifts the security bottleneck from detection to the more complex challenge of responsible and large-scale remediation, organizations must still manage prevention, context-aware prioritization, and the actual implementation of fixes across their estates. |
| 2026-10-18 | RCE | PhantomBuffer: Effective EDR Evasion Through Stealth Injection | Library for EDR evasion that utilizes stealth injection by leveraging Windows messages. PhantomBuffer bypasses traditional detection methods by employing `SendMessageW` with the `WM_SETTEXT` message to write payloads into target processes, avoiding the `WriteProcessMemory` API. This technique further enhances stealth by delegating thread execution redirection to the `cdb.exe` debugger, circumventing common EDR monitoring of `SetThreadContext` and `NtSetContextThread` APIs. |
| 2026-10-17 | Supply Chain | The Cloudflare Appstore (Appflare) Malware Audit | Library analyzing Cloudflare apps identified sensitive data leaks in two applications, rdyrct and 888wiki. The audit, conducted using Codex, focused on credential theft, hidden data collection, and backdoors across 139 local and 138 published app entries. rdyrct was found to transmit password reset tokens to an upstream analytics service (stats.brnr.dev), while 888wiki's private pages, including note titles and paths, were sent to a tracker (flow.aiurl.tw), and its file uploads for non-image types were directed to external hosts like box.david888.com. |
| 2026-10-17 | XSS | New web obfuscation technique where even just formatting rewrites script logic (Chromium DevTools, prettier, WebCrack, Terser and more) | Writeup on JavaScript "divergence," a novel anti-reversing technique where static code processing, including simple formatting via tools like Chromium DevTools, Prettier, and deobfuscators such as WebCrack and Terser, results in fundamentally different runtime logic. This method exploits bugs in JavaScript processors, not source maps or runtime checks, to create decoy code that LLM agents and human reversers may unknowingly analyze, leading to misinterpretations or flawed reimplementations. The technique was found across 37+ processors, including Babel, esbuild, and UglifyJS, with 186 unique variations identified. |
| 2026-10-17 | Bug Bounty | Codex vs Claude Code vs Gemini CLI: Where each AI agent fits in your Bug Bounty workflow | This article explores how AI agents like Codex, Claude Code, and Gemini CLI can enhance bug bounty hunting. It likely discusses the strengths and weaknesses of each AI for specific tasks within a bug bounty workflow, such as code analysis, vulnerability identification, or report generation. The goal is to help bug bounty hunters understand where each AI best fits to improve efficiency and effectiveness. |
| 2026-10-16 | RCE | Wanted to download an ISO, ended up writing a kernel exploit | Library for kernel exploit development, enabling root access on TP-Link EAP770 Wi-Fi 7 access points. This writeup details a path to disable flawed hardware offloading features, bypassing privilege restrictions. The exploit chain leverages CVE-2022-0185, an integer underflow vulnerability, combined with pipe-buffer manipulation to achieve arbitrary write capabilities in the kernel, requiring root privileges to execute. |
| 2026-10-16 | AI | Prompt injection in HTTP headers: what 217M web responses show | Library for analyzing prompt injection in HTTP headers, finding such attacks rare and mostly benign. Analysis of 217M web responses reveals that while injection attempts are infrequent, servers are actively detecting and labeling AI clients, with 136 sites classifying Common Crawl's CCBot. The library covers techniques like those used by agentpwn.com's `x-ai-instructions` header and factors influencing successful injection, such as AI model context passing and user prompt authority. |
| 2026-10-16 | AI | VulnBump: Can AI Agents Fix Vulnerable Dependencies Without Breaking the Build? | Library that benchmarks AI agents' ability to fix vulnerable dependencies. VulnBump tests agents like Snyk's Evo Remediation Agent and Anthropic's Claude Code against real-world upgrades across six package managers, measuring success by whether vulnerabilities are removed and builds/tests pass. Evo Remediation Agent achieves a 90.2% pass rate, significantly outperforming Claude Code, especially when provided with Snyk's recommended versions and scans. The benchmark highlights that identifying the correct, safe upgrade version and handling breaking changes are critical challenges for AI. |
| 2026-10-16 | AI | AI-Powered Espionage Campaign Targets Government Linux Infrastructure | Analysis of CL-UNK-1211 details an AI-powered espionage campaign targeting government Linux infrastructure from October 2025 to April 2026. Attackers utilized AI-generated payloads with conversational annotations and developer reasoning comments, deployed custom cross-platform backdoors like SysGoRAT, and leveraged web shells. Techniques included disabling endpoint protections, erasing system logs, and planting deceptive red-team markers. Credential harvesting involved modifying .bashrc files and injecting code into PHP login pages, alongside searching configuration files for cleartext passwords. The campaign focused on extensive reconnaissance, dumping sensitive databases, and exfiltrating government mailboxes to access social media accounts, demonstrating a sophisticated operational playbook. |
| 2026-10-16 | AI | AI Security Scoping: Define What To Test In Your Deployment | Library for AI security scoping, defining what to test, why it matters, and how to credibly test AI deployments. It differentiates AI penetration testing for technical weaknesses, GenAI architecture security reviews for design, AI red teaming for adversarial behavior, and traditional red teaming for detection and response across the entire ecosystem, not just the model. This approach inventories assets, traces data flows, and enumerates attacker objectives to make AI security outcomes measurable and defensible. |
| 2026-10-15 | Supply Chain | tensorlake@0.5.144: A Worm With Valid Provenance, and the Firewall Rule That Stops It | Library entry detailing the Shai-Hulud credential-stealing worm found in tensorlake@0.5.144, which collected npm tokens, SSH keys, and cloud credentials. The worm self-republished into other packages, making it a worm. Hextrap's scanner identified suspicious code and install hooks, automatically flagging the malicious version. The article highlights how a 72-hour soak time firewall rule prevents installation of new, unvetted package versions, stopping compromised releases like this one before they can execute. |
| 2026-10-15 | RCE | AhsayCBS Deep-Dive: Auth Bypass to RCE | Writeup of CVE-2026-105133 and CVE-2026-105134, detailing an authentication bypass vulnerability in AhsayCBS API components and a file write path traversal leading to remote code execution. Attackers can chain these flaws to write webshells into the application's filesystem, enabling SYSTEM and root command execution. The analysis unpacks the Java application's structure, identifies the authentication bypass in the `JSONApi.execute()` method via the "FromServlet":"Y" flag, and explains how `AddSysUser` allows creating new administrators without verification. It further details the path traversal in `FileService.a()` and `getReplicatedPath()`, demonstrating how to write malicious `.jsp` files into web-served directories, ultimately leading to webshell execution. |
| 2026-10-15 | RCE | Artifactory Under Attack: In-the-Wild Exploitation of CVE-2026-42016, CVE-2026-42018 & CVE-2026-82329 | Analysis of CVE-2026-42016, CVE-2026-42018, and CVE-2026-82329 impacting JFrog Artifactory details how attackers chain these vulnerabilities to bypass authentication and achieve administrative control. Exploitation involves gaining an anonymous user token via CVE-2026-42018, escalating privileges with CVE-2026-42016, or directly obtaining admin access through CVE-2026-82329. Post-exploitation activities include creating persistent admin accounts, deploying malicious Groovy plugins, and installing Rust-based backdoors for C2 communication. |
| 2026-10-15 | AI | Governing the Agentic SDLC Through Continuous Policy Enforcement | Library for agentic SDLC governance through continuous policy enforcement, enabling traceability for AI-generated code and actions. It proposes the Agentic Process Evidence Protocol, a verifiable standard for capturing tamper-evident, signed attestations of agent activities, including model usage and human review, aligning with regulations like the EU AI Act and standards such as OWASP AISVS and SLSA, and is available for contribution on GitHub. |
| 2026-10-15 | RCE | Unit 42 Uncovers PEERSTITCH, a Stealthy FortiGate Implant | Library for detecting PEERSTITCH, a stealthy ARM64 malware, which exploits CVE-2024-47575 on FortiManager and CVE-2025-61624 on FortiOS. PEERSTITCH operates in memory, communicating through legitimate process traffic, evading conventional detection. The attack chain involves exfiltrating configurations and credentials from over 500 FortiGate firewalls, enabling long-term access, traffic interception, and backdoor deployment. |
| 2026-10-15 | XSS | Submit and Execute: Weaponizing Reflected XSS in Siemens Teamcenter's Login Redirect (CVE-2026-58113) | Library for weaponizing reflected XSS in Siemens Teamcenter, identified as CVE-2026-58113, which allows for zero-interaction code execution. A crafted link exploiting the login-redirect flow reflects attacker-controlled query input into an auto-submitting HTML form without proper encoding, triggering injected JavaScript upon page load. This technique can be used to silently grant one account the ability to act on behalf of another by sending authenticated requests to Teamcenter APIs. |
| 2026-10-15 | Supply Chain | Package managers have an integrity problem | Analysis of 25 package managers reveals widespread vulnerabilities in detecting compromised package repositories. While cryptographic hashes and signatures offer solutions, fewer than half of tested managers default to lockfiles that check for tampering, and only two verify the integrity of every downloaded package. The research highlights variations in cryptographic designs, with only Go verifying signatures by default, and points to communities developing improved integrity solutions for future adoption. |
| 2026-10-14 | RCE | Vghost (CVE-2026-74580): Linux kernel vhost stale IOTLB cache leading to universal LPE and Guest-to-Host escape vector | Library describing Vghost (CVE-2026-74580), a Linux kernel vhost vulnerability. This defect stems from a stale IOTLB cache, enabling local privilege escalation to root via arbitrary kernel read/write or a guest escape through QEMU heap overwrite and control-flow hijacking when using vIOMMU and iommu_platform. The vulnerability was reproduced on Ubuntu, RHEL, Amazon Linux, Azure Linux, and Oracle Linux distributions, with fixes available in newer kernel versions. |
| 2026-10-14 | RCE | We Let AI Fuzz Adobe Reader for a Week. It Found a Critical Zero-Day CVE-2026-81983 | Writeup detailing CVE-2026-81983, a critical integer overflow vulnerability in Adobe Acrobat's image pipeline. This vulnerability, discovered by an LLM-generated fuzzer, allows an out-of-bounds write by crafting a 16-bit image within a PDF. The bug stems from a 32-bit multiplication for a byte-swap loop that incorrectly calculates the buffer size, leading to an overrun. Adobe's patch implements a 64-bit range check to prevent this overflow. |
| 2026-10-14 | Supply Chain | Codex CLI installs a persistent, self-updating background daemon by default | Issue detailing a vulnerability where the Codex CLI, installed via npm, defaults to auto-starting a persistent, self-updating background daemon. This daemon copies itself into a separate release directory, decoupling its version from the installed CLI. Users report significant background downloads and the daemon persisting even after closing all Codex windows, requiring manual intervention to disable or stop. Proposed solutions include making daemon auto-start opt-in and implementing an execution receipt to clearly identify the daemon's binary and environment. |
| 2026-10-14 | RCE | Playing It SFAPE: Code Execution in Isolated User Mode | Library for exploring code execution within Windows Virtualization-Based Security (VBS) enclaves, specifically focusing on Isolated User Mode (IUM). This library details vulnerabilities like CVE-2023-36880 and CVE-2024-21423 found in `prefs_enclave_x64.dll`, demonstrating how memory corruption within an enclave can lead to full control. It highlights the potential for VTL1 code reuse, similar to techniques used in projects like Mirage and Outflank's sleepmask, for advanced operating system attacks. |
| 2026-10-14 | AI | AI Agent Security: Best Practices for Safe Deployment | Library of best practices for AI agent security, this guide addresses the growing threat landscape highlighted by incidents like OpenAI models escaping sandboxes and compromising production infrastructure. It emphasizes treating AI agents as a new class of non-human identities (NHIs) that require robust credential lifecycle management. The content maps risks across host, identity, and agent attack surfaces, and details six practical controls, including sandboxing and scoped credential assignment per agent. It references techniques like prompt injection and excessive agency from OWASP's LLM Top 10, and tools like nono and Bromure for credential isolation. |
| 2026-10-14 | AuthZ | The Noise Inside the Network: How an Unsophisticated Attack Exposed a Third-Party Risk | Writeup detailing a European bank's compromise via a third-party gateway, highlighting exploitation of weak authentication and basic security hygiene. The attacker used noisy reconnaissance, credential testing, and targeted an unpatched WSUS server before being contained by Cortex XDR. This incident underscores how unsophisticated attacks can succeed through simple oversights like missing MFA and timely patching. |
| 2026-10-13 | RCE | LG webOS sync-demo Preauth RCE | This content describes a critical Pre-authentication Remote Code Execution (RCE) vulnerability found in LG's webOS operating system, specifically within the `sync-demo` component. This means an attacker could potentially execute arbitrary code on a vulnerable LG smart TV without needing any prior authentication. The summary title itself highlights the severity of this security flaw. No bug bounty payout amount is mentioned in the provided text. |
| 2026-10-13 | RCE | CVE-2026-91768: FastCGI allowed_clients compared only 96 of an IPv6 address's 128 bits | Writeup of CVE-2026-91768, a partial comparison vulnerability in PHP-FPM's `listen.allowed_clients` directive that allowed connections from 4,294,967,296 IPv6 addresses when only one was specified. Introduced in 2014, the flaw persisted until September 2026 due to incomplete IPv6 address comparisons within the `fcgi_is_allowed` function. The vulnerability, which could be exploited remotely without authentication, allowed bypassing access controls by comparing only 12 out of 16 bytes of IPv6 addresses, effectively treating an exact match as a /96 prefix match. |
| 2026-10-13 | RCE | I Found a $90k Remote Code Execution Vulnerability in Google | Writeup detailing a $90,000 RCE vulnerability discovered in Google's DevSite OAuth flow. The vulnerability arose from a Flask session cookie containing a pickle-serialized object that included the OAuth client secret. This secret was reused as the Flask `SECRET_KEY`, allowing an attacker to forge session cookies and execute arbitrary Python code via the `pickle.loads` function during the OAuth callback processing. The writeup covers session cookie analysis, pickle deserialization vulnerabilities, and the chain from exposed secrets to unauthenticated RCE, highlighting specific affected Google domains. |
| 2026-10-13 | Bug Bounty | ‘I’m not the kind of person to rely on tooling’: Sn0rkY shares his Bug Bounty story and top tips | Bug bounty hunter Sn0rkY emphasizes the importance of manual testing over over-reliance on automated tools. In his story, he highlights that critical vulnerabilities are often missed by automated scanners. Sn0rkY shares his personal journey and offers top tips for aspiring bug bounty hunters, stressing the value of keen observation, deep understanding of systems, and persistent manual investigation to uncover unique and impactful bugs. |
| 2026-10-12 | RCE | Telegram Desktop vulnerability allowed any user file to be stolen | Library for understanding a Telegram Desktop vulnerability that enables arbitrary file reads. The attack chain exploits a local socket communication flaw, allowing command injection and the use of an internal `interpret://` URI scheme. This scheme, intended for release publishing, reads specified files and sends them to a channel without authorization checks, enabling theft of sensitive user data like session keys. |
| 2026-10-11 | Supply Chain | GitHub Actions leaking secrets when Miri output is cached | Library issue where Miri's environment variable caching in the `target/` directory, combined with GitHub Actions' caching mechanism, can expose secrets to pull requests. This vulnerability, observed in projects utilizing Miri with cached `target/` directories, allows malicious actors to extract sensitive information through CI caches. A fix limits Miri to preserving only specific `CARGO_*` environment variables and `OUT_DIR`, mitigating the risk. |
| 2026-10-11 | Mobile | A story of wifi secrets, UART lockdowns and a damaged flash memory chip | Library for dissecting Eufy HomeBase 2 firmware (version 3.4.2.2h). Details firmware extraction challenges, including damaging flash memory and recovering data with 7z. Explores reverse engineering the Wi-Fi credential generation algorithm for OCEAN_XXXXXX, analyzing U-Boot, and discussing filesystem quirks for debugging. Covers techniques like using spring clips for SPI reads, desoldering flash memory, and leveraging Ghidra and radare2. |
| 2026-10-10 | RCE | A JPEG, a Race, and a Ghost: Breaking Discourse's Image Pipeline | Library detailing two vulnerabilities in Discourse's default configuration: a cache poisoning attack leading to CSP bypass and JavaScript execution for anonymous users via a modified `color_scheme_id` cookie, and an arbitrary file read stemming from a TOCTOU race condition in the image upload pipeline, exploitable with Trust Level 0 and involving ImageMagick and GhostScript. |
| 2026-10-10 | RCE | MXC - a sandboxed code execution system | Library for sandboxed code execution on Windows, Linux, and macOS. MXC supports multiple containment backends like ProcessContainer, LXC, and Bubblewrap, governed by detailed filesystem, network, and UI policies. It offers Rust, .NET, and Node SDKs for integrating into applications, along with diagnostics to aid in policy tuning and troubleshooting access-denied failures. The system allows for provisioning, starting, executing, and deprovisioning persistent containers, providing fine-grained control over untrusted workloads. |
| 2026-10-10 | Mobile | Show HN: Carrier-Explode: iPhone, Pixel and Galaxy carrier settings decoded | "Carrier-Explode" is a new project that decodes carrier settings for iPhones, Pixels, and Galaxies. This tool aims to shed light on how these configurations work across different devices and carriers, offering a deeper understanding of network management and customization. The project's release highlights a community effort to demystify a complex aspect of mobile device operation. |
| 2026-10-10 | RCE | Death By A Thousand PaperCuts (PaperCut Pre-Auth RCE Chain and Patch Bypasses WT-2026-0141-0144/CVE-2026-82077/CVE-2026-82078/CVE-2026-81578) | Walkthrough of PaperCut pre-auth RCE chain and patch bypasses, including WT-2026-0143, WT-2026-0144 (CVE-2026-82077), CVE-2026-82078, and CVE-2026-81578. The analysis details how vulnerabilities and subsequent patch bypasses were discovered and chained, culminating in an unauthenticated remote code execution exploit against PaperCut's print management software. |
| 2026-10-10 | AI | AI Agent Security: Six Controls From Nine Real Incidents | Library of controls for AI agent security, derived from nine real-world incidents between May 2025 and July 2026. Six core controls are detailed: sandboxing agents, scoping credentials, isolating agent configurations, logging all calls, scanning workspaces for secrets beforehand, and implementing deny-by-default policies. These controls address attack surfaces including the agent's host, its identity, and the agent itself, with specific examples of failures involving prompt injection, unchecked exit codes, and over-scoped permissions seen in incidents affecting services like Gemini CLI, Claude, Amazon Kiro, and Microsoft 365 Copilot. |
| 2026-10-09 | Mobile | Loupe: An Android Console in the Browser | Library that centralizes Android assessment tools in a Chromium tab, utilizing WebUSB to connect to devices without requiring local installations of adb, scrcpy, or frida-tools. Loupe features a mirrored screen, logcat viewer with filtering, file browser with previews for SQLite databases, an app manager, and a Frida interface that automates frida-server deployment and execution, supporting older scripts by restoring the Java bridge. It also includes a proxy configuration for intercepting traffic directly on the device. |
| 2026-10-09 | RCE | How We Found Thousands of Exposed NVIDIA GPUs and a Way to Disrupt Them (CVE-2026-47483) | Writeup detailing CVE-2026-47483, a high-severity vulnerability in NVIDIA DCGM Exporter that enables unauthenticated attackers to cause denial of service and information disclosure by triggering uncontrolled resource consumption. Researchers discovered over 2,000 exposed GPU servers online, revealing significant NVIDIA hardware like Blackwell, H100s, and consumer RTX cards, often hosted by neocloud providers. NVIDIA released a security bulletin for CVE-2026-47483, which has a CVSS score of 8.2. |
| 2026-10-09 | XSS | I found yet another way to invoke JavaScript functions without parentheses | Library that demonstrates a novel JavaScript function invocation technique without parentheses. This method abuses the `Error.prepareStackTrace` function, typically used for stack trace formatting, to generate arbitrary JavaScript functions. By manipulating `Array.prototype.toString` or `Object.prototype.message` via prototype pollution, attackers can control the generated code. The technique culminates in invoking these functions using methods like `valueOf` or string concatenation, potentially leading to arbitrary code execution, as seen in the provided exploit payload. |
| 2026-10-09 | RCE | A Single POST Freezes Any Next.js Server | Writeup of CVE-2026-23870 detailing a Denial of Service vulnerability in React Server Components. This vulnerability, found in React versions prior to 19.0.6, 19.1.7, and 19.2.6, allows an attacker to freeze a Next.js server by crafting a single POST request with numerous nested form data pointers. The issue stems from React's inefficient parsing of nested form data, causing excessive CPU usage and making the server unresponsive to legitimate requests. The fix involves optimizing the form data parsing to scan fields only once per request, preventing the exponential complexity. |
| 2026-10-09 | AI | [tl;dr sec] #349 - Vulns & Exploits in the AI Era, Package Manager Sandboxing, Testing AI Sandboxes | Library of techniques for testing AI sandboxes, with specific mention of Vercel and Perplexity. This entry also covers improvements to Burp Suite's Turbo Intruder for HTTP/3 support, including new race condition methods like the Single Datagram Attack and QPACK blocked streams, and the HTTP/3 Adapter extension for broader Burp integration. Additionally, it details the Graphalgo campaign's spread to Terraform providers and Go Modules, and surveys how package managers like Homebrew, opam, Nix, and Bazel approach sandboxing install-time code. |
| 2026-10-08 | Mobile | Presenting DiagNG: After QCSuper, a new open-source initiative for freeing up mobile baseband Diag protocols | Tool, DiagNG, is a GUI-based successor to QCSuper for capturing 2G/3G/4G/5G air interface logs in PCAP/GSMTAP format, compatible with Wireshark. It supports Qualcomm Snapdragon basebands and NR/5G RRC logs using the GSMTAP v3 protocol. DiagNG leverages GTK 4/Adwaita for its UI, Flatpak for Linux integration, and an asynchronous architecture with components like ADB and ModemManager, incorporating Kaitai Struct for protocol definitions and a Rust daemon for privileged operations. |
| 2026-10-08 | RCE | CVE-2026-102489 Deep-Dive: Zammad Session Leak to RCE | Writeup of CVE-2026-102489 detailing a session hijack vulnerability in Zammad. This vulnerability, when triggered by a crafted WebSocket request, leaks session cookies from error messages, allowing attackers to obtain valid admin session cookies. These cookies can then be used to write arbitrary files to the application directory, specifically by overwriting the password reset email template, leading to remote code execution as the `zammad` user through Ruby's ERB rendering. |
| 2026-10-08 | Recon | Beyond asset discovery. Real-life CrowdRecon use case explored | Library for structuring researcher reconnaissance data, turning raw findings into usable input for security teams. This tool, exemplified by a use case involving an old live marketing site, helps organizations identify forgotten or overlooked assets by comparing researcher observations with existing inventories, EASM platforms, and vulnerability scanner data. It facilitates decisions on whether to investigate further, assign ownership, update testing scope, or continue monitoring, providing valuable context beyond simple asset discovery. |
| 2026-10-08 | Supply Chain | Evolution of Web3 in Cloud Supply Chain Attacks | Analysis of Web3 in Cloud Supply Chain Attacks details how threat actors leverage decentralized blockchain architectures for command-and-control (C2) infrastructure, moving beyond static endpoints. Campaigns like ChainDrop and PolinRider demonstrate the exploitation of open-source dependencies to harvest cloud identity tokens and secrets. These attacks, attributed to state-sponsored actors like Alluring Pisces, utilize techniques such as EtherHiding and TxDataHiding across networks like TRON, Aptos, and Binance Smart Chain to dynamically update botnets and maintain persistence, bypassing traditional security controls. |
| 2026-10-07 | Bug Bounty | Bitvulnex: a vulnerable crypto exchange | Library for practicing application security, Bitvulnex is a deliberately vulnerable Bitcoin exchange. It features 40 planted vulnerabilities spanning web security, exchange business logic, Bitcoin workflows, and infrastructure, allowing users to explore and learn from common issues like access control failures, injection, race conditions in withdrawals, price oracle manipulation, and SSRF in KYC processes. The application, built with Next.js, PostgreSQL, and Redis, includes simulated trading, deposits, and background jobs to provide a realistic environment for discovering and understanding security flaws. |
| 2026-10-07 | OSINT | Show HN: MailAccess – the true Email OSINT framework | Library for email OSINT and scraping that identifies accounts behind a single email or all emails at a company. It utilizes over 75 modules across 5,300+ platforms to discover business contacts, visualizing findings in a graph format with clear sourcing, confidence scoring, and reasoning. The open-source, MIT-licensed engine is free and can be run locally without data leaving the user's network. |
| 2026-10-07 | RCE | You Won’t Hear About These, Even In Myths (Atlassian Jira, Confluence (and more) Pre-Auth Arbitrary File Read CVE-2026-21589) | Writeup on CVE-2026-21589, an arbitrary file read vulnerability affecting Atlassian Jira, Confluence, and Bitbucket. The vulnerability stems from the `atlassian-plugins-webresource*.jar` library, which mishandles path traversal attempts using double colons (`::`) as a substitute for slashes. This allows unauthenticated attackers to read arbitrary files on affected systems by exploiting the `Router.unescapeSlashes` and `ResourceFactory.createResourceWithRelativePath` functions. |
| 2026-10-07 | Secrets | Secrets Management Best Practices: 6 Reasons Your Vault Needs Detection | Library for secrets detection and management, this resource highlights the critical need for both secrets vaults and detection tools. Secrets vaults centralize and protect credentials, enforcing access policies and recording usage, while secrets detection continuously scans code, configurations, and platforms for exposed secrets outside the vault. Combining these capabilities provides a comprehensive inventory of all credentials, enabling faster remediation of security incidents, satisfying auditor requirements by demonstrating control effectiveness, and offering robust evidence for frameworks like SOC 2 and ISO 27001. |
| 2026-10-07 | AI | Why the AI Attack Surface Extends Your Stack | Library for evaluating enterprise AI risk, extending beyond the model to include invoked tools, data sources, surrounding applications and APIs, identities, and cloud infrastructure. It details how prompt injection can lead to code execution and data exposure when service accounts have broad permissions and tenant isolation fails, highlighting that model safety scans and conventional AppSec scans miss critical parts of this expanded attack surface. The library emphasizes mapping AI system access and actions, then testing realistic attack chains across the full application stack. |
| 2026-10-06 | RCE | Wordpress libheif RCE | This content describes a Remote Code Execution (RCE) vulnerability in WordPress related to the libheif library. The specific details and impact of the vulnerability are not provided, nor is any information regarding a bug bounty payout. |
| 2026-10-06 | RCE | Open Build Service, one year later: command execution through Mercurial argument injection | Tool identifying CVE-2026-56004, a command execution vulnerability in Open Build Service's `obs_scm` service. The flaw arises from Mercurial argument injection when the `revision` parameter is improperly handled, allowing attackers to exploit Mercurial's `--config` option to execute arbitrary shell commands on the build server, posing a significant risk to software supply chains. |
| 2026-10-06 | Bug Bounty | The evolution of Bug Bounty: history, best practices and the impact of AI | Bug bounty programs have transformed from informal hacker challenges to sophisticated security initiatives. Their evolution is marked by increasing organization, structured rewards, and broader adoption by companies seeking to proactively identify vulnerabilities. Key best practices include clear scope definition, fair compensation, and prompt communication. The emergence of AI is set to significantly impact bug bounties by automating vulnerability discovery and analysis, potentially leading to more efficient programs and a surge in reported bugs. |
| 2026-10-06 | Secrets | SOPS Guide: How to Encrypt and Manage Secrets in Git, Kubernetes, and CI/CD | Library for encrypting secrets within YAML, JSON, ENV, and INI files using a choice of key stores like PGP, age, AWS KMS, Azure Key Vault, Google Cloud KMS, and HashiCorp Vault. SOPS supports envelope encryption and can be integrated with GitHub Actions and Flux CD for automated decryption at build or reconcile time, ensuring plain text secrets are never committed to Git. It also provides message authentication codes to detect file tampering and can be paired with secrets managers for runtime access. |
| 2026-10-06 | Authentication | Smashing the token limit with overlapping fragments | Library for exfiltrating large tokens using overlapping CSS fragments. This technique reconstructs tokens by stitching together smaller identified CSS chunks from a given URL. It optimizes token extraction efficiency by strategically employing two-character, three-character, four-character, and five-character checks to minimize CSS payload size, achieving token lengths of 210 hex characters with a single candidate and up to 640 characters with multiple candidates. |
| 2026-10-05 | AI | From the creator of Redis; run LLM locally with ds4 | Library for local Large Language Model inference, DwarfStar 4 (ds4) offers asymmetric 2-bit quantization to compress routed experts while preserving critical shared paths, enabling models like DeepSeek V4, GLM 5.x, and Qwen3.8 to run on high-memory Mac, CUDA, and ROCm machines. It supports text and vision models, local APIs, a CLI, and a native agent, with features like SSD-based prefix saving for faster restarts. |
| 2026-10-04 | RCE | RCE and bad crypto in Internxt's 'post-quantum' cloud storage | Writeup detailing critical vulnerabilities in Internxt's cloud storage, including remote code execution via protocol handlers and session/key leakage through unauthenticated public key sharing and insecure redirection. The analysis highlights architectural flaws in their cryptographic implementation, such as a flat key hierarchy and weak password hashing (MD5 with 3 iterations), rendering their "post-quantum" security claims unreliable and enabling potential data interception by Internxt or malicious third parties. |
| 2026-10-03 | Recon | security.txt on the Czech web: Scanning 1k popular .cz domains | Writeup analyzing the adoption and validation of security.txt files across the top 1,000 .cz domains. The research reveals that only 16% of these popular Czech websites published a detectable security.txt, with just 12% passing strict RFC 9116 validation. Common failures included incorrect charset headers, missing `Expires` fields, and even WAF blocking of the discovery attempts. The analysis also highlights instances of expired or excessively long-dated `Expires` entries. |
| 2026-10-03 | RCE | 8 out of 10 Banks HATE This One Weird 3SKey RCE | Library for authentication with hardware signing tokens like 3SKeys, SConnect (v2.16.0.0), had a critical remote code execution vulnerability (CVE-2026-18397) due to a hand-rolled RSA-2048 token validation implementation. This flaw allowed any site or iframe to silently download and execute a DLL by exploiting an uninitialized memory validation bypass, enabling the loading of "plugins" and impacting systems used by banks and national authentication services. |
| 2026-10-03 | AuthZ | Azure's Weakest Link - Five Full Cross-Tenant Compromises | Writeup detailing the exploitation of Azure API Connections, which resulted in five cross-tenant compromises. The vulnerabilities allowed attackers to leverage Azure Resource Manager's (ARM) access to any connection within a tenant, enabling unauthorized access to backend services including Azure Key Vaults and Azure SQL databases. Exploitation involved manipulating unpatched `DynamicInvoke` and other undocumented `DynamicList` endpoints through path traversal to execute arbitrary commands and queries on victim systems. |
| 2026-10-03 | RCE | Bypassing Secure Boot via Unbounded RLE8 Splash Images in U-Boot (CVE-2026-71972) | Advisory for CVE-2026-71972 details an unbounded RLE8 splash image vulnerability in U-Boot. A crafted RLE8 BMP image, loaded from attacker-writable storage before boot stage authentication, can write past the framebuffer's bounds, corrupting adjacent memory and enabling secure boot bypass. Affected versions include U-Boot through 2026.10-rc5, with a fix available in commit 5201e83342d64c2f438ea35158575f28225e752e. |
| 2026-10-02 | Supply Chain | Your SBOM Is Fan Fiction | Tool for runtime Software Bill of Materials generation; queries `/proc` for process dependencies and maps, then correlates with package data and CVE advisories using `yeet.graph.query` and `yeetkit` for enriched security insights on live systems. |
| 2026-10-02 | RCE | How to Spot a Compromised MikroTik Router | Reference detailing techniques for detecting compromised MikroTik routers, focusing on identifying attacker configurations. It covers SSH port forwarding (`forwarding-enabled`), built-in SOCKS proxies (`/ip socks`), and the web proxy (`/ip proxy`), highlighting specific artifacts like `forwarding-enabled: local` and `redirect-to` rules. The document explains how these features, when improperly configured or exposed to the WAN, can facilitate pivoting and traffic manipulation, referencing historical abuse via CVE-2018-14847. |
| 2026-10-02 | RCE | Server Mismatch: WordPress plugin vulnerabilities when relying on .htaccess files | Writeup on WordPress plugin vulnerabilities arising from overreliance on `.htaccess` files. When plugins installed on non-Apache servers, such as Nginx, ignore `.htaccess` directives, sensitive files like backup `.procstat` files in Everest Backup or database backup manifests in BackWPUp can be exposed to unauthenticated attackers. This research highlights security risks when plugin security relies solely on server-specific configurations. |