Supply Chain Security
Software supply chain security addresses threats that target the dependencies, build systems, and distribution channels that modern applications rely on. High-profile incidents like SolarWinds, Log4Shell, and the xz backdoor demonstrated that attackers increasingly target upstream components rather than applications directly. Supply chain attacks include dependency confusion (substituting malicious packages with names matching internal packages), typosquatting in package registries, compromised maintainer accounts, malicious code injected into build pipelines, and trojanized development tools. Defenses include software bills of materials (SBOMs), dependency pinning and lock files, signature verification, provenance attestation (SLSA framework), regular dependency auditing with tools like Dependabot, Snyk, or Socket, and careful evaluation of new dependencies before adoption.
You audit your code and ship everyone else's
A typical application is a small amount of first-party code on top of a very large dependency tree, built by a pipeline with broad credentials and deployed through infrastructure that trusts its own artifacts. Reviewing the first-party code carefully while treating everything underneath as given is the normal state of affairs, and it is the gap this category is about.
The attack surface splits usefully into three. Package ecosystems: typosquatted names, dependency confusion where a public registry answers for an internal package name, compromised or transferred maintainer accounts, and install-time script execution that runs before anyone has read anything. Build systems: CI configuration is code with credentials, and a pull request that triggers a privileged workflow has historically been enough to extract them; caches and artifact stores are tempting because poisoning one affects every consumer downstream. Distribution: signing key compromise, update channel manipulation, and container base images that nobody has rebuilt.
What makes these attractive is amplification. Compromising one widely-used package reaches every application that installs it, which is a return no application-level bug offers. The incidents that shaped current practice — SolarWinds, Codecov, event-stream, the recurring npm and PyPI campaigns — are worth reading in full rather than in summary, because the interesting detail is usually how long the access went unnoticed and what specifically failed to detect it.
On defence, the honest position is that the ecosystem is still building the primitives. SBOMs give you an inventory, which is necessary and not sufficient. Pinning with hash verification prevents silent substitution. Reproducible builds and artifact signing with frameworks like SLSA and Sigstore address provenance. Restricting install-time scripts and isolating CI credentials are unglamorous and effective. Vendoring trades update friction for control.
The practical starting point is inventory: most organizations cannot currently answer which applications include a given package version, and everything else depends on being able to.
| Date Added | Link | Excerpt |
|---|---|---|
| 2026-11-20 NEW 2026 | Hardening GitHub Actions Against Insider Threats beginner 38 min read | Library for hardening GitHub Actions against insider threats, focusing on stolen credentials. It details controls for reducing the blast radius of compromised accounts, such as requiring phishing-resistant factors like WebAuthn security keys, enforcing authentication at the organization level, shortening session lifetimes, treating recovery codes as secrets, and narrowing the publish path with techniques like trusted publishing and staged releases. It also addresses vulnerabilities like device-code phishing and the limitations of 2FA against personal access tokens and SSH keys. → socket.dev |
| 2026-11-20 NEW 2026 | Microsoft’s Agent Framework PyPI package compromised to steal credentials news 2 min read AI Python | Library compromise involving Microsoft's Agent Framework on PyPI saw malicious versions 1.26.1 and 1.26.2 published on November 17, 2026, to steal cloud credentials. The malware, executed during Python startup, targeted Azure, AWS, GitHub, and OpenAI secrets, including obtaining Azure managed identity and GitHub Actions OIDC tokens. Discovery by Socket, Step Security, and Ox Security led to PyPI quarantining the releases, but the attacker quickly re-published. Developers who installed these versions should isolate systems and rotate all potentially exposed credentials. → bleepingcomputer.com |
| 2026-11-20 NEW 2026 | The Tampered Claude Code Extension in Cursor’s Marketplace: What It Did and How to Check news AI | A tampered Claude Code extension in Cursor's marketplace planted an SSH backdoor for approximately eleven hours on October 29th. This malicious code allowed unauthorized access to user systems. Users are advised to check their installations and remove any suspicious versions of the extension to mitigate potential security risks. |
| 2026-11-20 NEW 2026 | The Fireworks AI Breach: How One Vendor Key Became a Supply Chain Attack on Cursor news 9 min read AI Secrets | Library for detecting leaked credentials on developer machines and preventing secrets from entering supported coding assistants, this resource details the Fireworks AI breach. The supply chain attack began with a read-only vendor key, leading to the compromise of Cursor's internal tools and the distribution of trojanized Claude Code extensions to potentially 5,000 developers. The breach also exposed Notion AI chat transcripts containing user-entered API keys and other sensitive information. → blog.gitguardian.com |
| 2026-11-20 NEW 2026 | One Text Box, 178 Organizations: How a Trusted Script Turned the BBC’s Homepage Into an Attack news XSS | A widely trusted third-party JavaScript file, utilized by 178 organizations including the BBC, Zoom, and CNN, was compromised. This allowed attackers to inject malicious code, effectively turning the BBC's homepage into an attack vector. While no direct breaches of these major companies occurred, the vulnerability exploited a shared dependency, demonstrating the significant risks associated with third-party script integrations. The incident highlights the interconnectedness of the digital ecosystem and the potential for widespread impact when a single trusted component is compromised. |
| 2026-11-19 NEW 2026 | Microsoft agent-framework PyPI Package Compromised news 8 min read Python | Library of malicious Python code discovered within the Microsoft agent-framework PyPI package, specifically versions 1.26.1 and 1.26.2. The compromised packages utilized `.pth` startup hooks to execute a credential-stealing payload upon Python interpreter startup. This payload, designed to exfiltrate environment variables, credential files, and mint identity tokens from GitHub Actions OIDC, Azure DevOps OIDC, and Azure managed identity endpoints, demonstrated rapid iteration with two malicious releases published within 33 minutes. → stepsecurity.io |
| 2026-11-19 NEW 2026 | Microsoft’s agent-framework PyPI package Compromised with Credential Stealer news 3 min read AI Python | Writeup detailing a credential stealer embedded in Microsoft's agent-framework PyPI package, specifically versions 1.26.1 and 1.26.2. These versions utilize Python startup hooks (.pth files) to exfiltrate Azure, GitHub Actions, and CI/CD credentials by scanning environment variables, local files, and cloud identity endpoints before sending them to a webhook.site URL. → socket.dev |
| 2026-11-19 NEW 2026 | Cursor Says Stolen Fireworks AI Key Led to Trojanized Claude Code Extension Incident news 3 min read AI Secrets | Incident report detailing the trojanized Claude Code extension distributed via Cursor Marketplace. The breach stemmed from stolen Fireworks AI credentials, leading to unauthorized access of Cursor's read-only storage, exposing customer support data, source code, and API credentials. Attackers then placed malicious versions of the extension, potentially affecting 5,000 users, which attempted to install SSH backdoors. Uber and Notion were also impacted by the underlying Fireworks AI compromise. → thehackernews.com |
| 2026-11-19 NEW 2026 | A Python Startup Hook, a Hijacked Microsoft Package, and 41 Minutes on PyPI news Python | On November 17, 2026, a malicious actor compromised the `microsoft` PyPI package. They uploaded a new version containing a backdoor that would execute a Python startup hook. This hook allowed the attacker to run arbitrary code on any system that updated to the compromised package. The malicious version was live on PyPI for 41 minutes before being discovered and removed by the Python Package Index security team. |
| 2026-11-19 NEW 2026 | Optimizely Supply Chain Attack Distributes Malware Across Global Brands news 3 min read | Analysis of the Optimizely Supply Chain Attack details how attackers exploited a vulnerability in Optimizely's Web Experimentation platform, injecting malicious JavaScript into global brands like BBC, CNN, and HSBC. This attack delivered cryptocurrency clippers, card skimmers, and iOS malware by compromising trusted third-party scripts. The vulnerability resided in the Project JavaScript field, allowing the execution of custom code, and was patched by Optimizely. → socket.dev |
| 2026-11-19 NEW 2026 | Hackers Found a Way Around GitHub SHA Pinning. This Webinar Explains How to Close the Gap intermediate 3 min read Talks | Webinar addressing how attackers bypassed GitHub SHA pinning, a security measure against supply chain attacks. The session covers how malicious code disguised as security fixes infiltrated Trivy and actions-cool/issues-helper, exploiting GitHub's fork sharing. It clarifies that while pinning ensures immutability, it doesn't guarantee authorship or project origin. Attendees will learn about the limitations of org-level pin enforcement, impostor commits, GitHub's Actions roadmap, and methods to close trust gaps, including a catalog of vetted actions, to secure CI/CD pipelines beyond basic pinning. → thehackernews.com |
| 2026-11-18 NEW 2026 | Two malicious agent-framework releases hit PyPI in 33 minutes news 10 min read Secrets | Library releases agent-framework 1.26.1 and 1.26.2, malicious versions published to PyPI, install a credential stealer. This stealer executes on every Python start, targeting secrets for Azure, Azure DevOps, and GitHub Actions. The attack exploits `.pth` files, a mechanism typically used to add directories to the import path, but executable if the line begins with `import`. Detection is available via Aikido Security's feed, filtering for malware issues. Remediation involves pinning the `agent-framework` package to version 1.26.0 and rotating all potentially compromised secrets. → aikido.dev |
| 2026-11-18 NEW 2026 | Borrowed Code, Real Consequences: What OPTIMJACK Teaches Us beginner 3 min read | Writeup on the OPTIMJACK supply chain attack, demonstrating how a vulnerability in Optimizely's Web Experimentation platform led to malicious JavaScript injection across 178 organizations. The compromised code, served via the vendor's CDN, facilitated cryptocurrency wallet swapping, payment card skimming, and exploit kit delivery. This incident underscores the risks of third-party scripts with dynamic execution privileges and highlights segmentation as a crucial defense mechanism, such as isolating payment collection within cross-origin iframes and removing non-essential tags from sensitive pages. → unit42.paloaltonetworks.com |
| 2026-11-18 NEW 2026 | Patch Me If You Can: Tracing CVEs to Deployed Code advanced 6 min read | Analysis of CVE-2026-111497, CVE-2026-109618, CVE-2026-109619, and CVE-2023-4863 demonstrates how application code analysis, rather than just version checks, is crucial for determining patch status. Experiments with Node.js and Brotli, PHP backports in Apache, and a separate libwebp installation via ImageMagick reveal that application dependencies and runtime environments can obscure vulnerability fixes, necessitating deeper inspection than package management alone provides. → bishopfox.com |
| 2026-11-18 NEW 2026 | 178 Organizations Served Malicious Code Through Trusted Websites. Here’s What You Should Check. news | A sophisticated supply-chain attack compromised 178 organizations, including major brands like BBC, CNN, and Zoom, by serving malicious code through the trusted platform Optimizely. Attackers exploited this vulnerability to inject harmful scripts, potentially affecting users who visited these compromised websites. The attack highlights the risks associated with third-party code and emphasizes the need for organizations to rigorously vet and monitor their software supply chains for vulnerabilities. |
| 2026-11-18 NEW 2026 | How to protect your JS/TS codebase from supply chain attacks beginner | Supply chain attacks on the NPM ecosystem are increasing, targeting JS/TS codebases. Developers need robust protection strategies. This content likely explores methods to safeguard projects from compromised dependencies, malicious packages, and other vulnerabilities that exploit the software supply chain. Protecting your codebase involves vigilance and proactive security measures within your development workflow. |
| 2026-11-18 NEW 2026 | Was my GitHub hit by wfx-loot? A 15-minute read-only check news | A campaign named "wfx-loot" recently targeted approximately 4,800 GitHub accounts, including both user and bot accounts, by exploiting stolen tokens. This campaign began on November 10th. The article offers a quick, 15-minute read-only method to check if your GitHub account was affected by this incident. |
| 2026-11-18 NEW 2026 | Two Shells, Two Poisoned Releases, and Five Controls You Can Start This Week news | October's supply chain chaos, illustrated by two "poisoned releases" impacting Two Shells, highlights critical security vulnerabilities. The article distills lessons learned from these incidents and offers five actionable controls that can be implemented immediately. It emphasizes practical security measures, sharing insights from a Docker Captain's approach to risk mitigation. |
| 2026-11-18 NEW 2026 | Hijacked Optimizely Scripts Served Card Skimmer and iOS Exploit Kits to CNN, BBC and Zoom Visitors news 7 min read Mobile | Library for detecting and mitigating attacks against Optimizely Web Experimentation, including card skimming, crypto wallet address swapping, and iOS exploit kits like Coruna and DarkSword. The library helps identify malicious JavaScript injected into customer projects, which affected sites like CNN, BBC, and Zoom. It aids in analyzing suspicious code patterns, reviewing user access, and understanding the impact of compromised session tokens and card details. → thehackernews.com |
| 2026-11-18 NEW 2026 | Microsoft agent-framework Compromised: Credential Stealer Targets Azure and CI Pipelines news 4 min read AI Python Secrets | Library versions 1.26.1 and 1.26.2 of Microsoft's agent-framework, published to PyPI, contained a credential stealer targeting Azure and Azure DevOps. The malicious package drops a `.pth` file that executes on interpreter start, stealing environment variables, credential files, and Azure/GitHub OIDC tokens. The stealer exfiltrates data via HTTPS POST to webhook[.]site, with limited impact due to the free tier rate limiting. → wiz.io |
| 2026-11-17 NEW 2026 | WPPConnect Compromised: Malicious Releases Steal Secrets and WhatsApp Sessions Through the Project's Own Pipeline news 10 min read Secrets | Library compromised by malicious releases of @wppconnect-team/wppconnect and @wppconnect/server, versions 2.4.1 and 2.10.82 respectively, which exfiltrated host environment variables, .env files, and WhatsApp session tokens to a Telegram bot. The attack exploited the project's own CI/CD pipeline, with malicious code running on import, bypassing `npm install --ignore-scripts`. Valid trusted publisher credentials and SLSA provenance were used, and the release tag was later rewritten on GitHub to obscure the payload. → stepsecurity.io |
| 2026-11-17 NEW 2026 | Cursor Supply Chain Attack Started at Fireworks AI news 2 min read AI | Writeup of the Cursor supply chain attack, which originated with a compromise at Fireworks AI. Threat actors leveraged stolen Fireworks credentials to access Cursor's environment, then distributed malicious copies of the Anthropic Claude Code extension via the Cursor VS Code Extension Marketplace. This incident impacted up to 5,000 users who installed compromised versions 2.1.312 or 2.1.313. → securityweek.com |
| 2026-11-17 NEW 2026 | Trojanized Claude Code Extension Delivered Persistent SSH Backdoor to Developer Machines news 6 min read AI | Writeup of trojanized Claude Code VS Code extension; versions 2.1.312 and 2.1.313, distributed via Cursor's marketplace, contained a malicious loader. This loader performed reconnaissance, implanted an SSH backdoor with an Ed25519 key, enabled SSH services, established a reverse tunnel, and implemented persistence mechanisms via cron, launchd, or scheduled tasks. The beacon stage reported on the success of these implants. The attack did not involve Anthropic's systems but targeted developer machines through tampered extension bundles. → stepsecurity.io |
| 2026-11-17 NEW 2026 | CVE-2026-126187: React Server Components Flaw Lets Attackers Execute Code Through Smuggled Flight Rows news 9 min read RCE | Writeup of CVE-2026-126187, an unauthenticated remote code execution vulnerability in React Server Components affecting react-server-dom-webpack, react-server-dom-parcel, and react-server-dom-turbopack. This critical flaw, scoring 9.5 on CVSS 4.0, arises from a deserialization of untrusted data (CWE-502) where attackers can smuggle malicious module rows into corrupted RSC responses, leading to arbitrary module execution. Fixes were released on November 10, 2026. → stepsecurity.io |
| 2026-11-17 NEW 2026 | Contentlayer code injection intermediate RCE | Advisory on GHSA-gxcv-hr94-jjmj detailing a code injection vulnerability in Contentlayer's @contentlayer/core through version 0.3.4. Attackers can exploit this by embedding single quotes in content file names, leading to JavaScript execution when contentlayer dev or a Next.js dev server imports the generated module. The vulnerability stems from an issue within the `makeDataExportFile()` function. |
| 2026-11-17 NEW 2026 | uBlock Origin’s Universal XSS Is Patched. A Filter It Planted Can Still Sit in Your My Filters Tab. intermediate XSS | uBlock Origin's universal cross-site scripting (XSS) vulnerability has been patched. This exploit allowed malicious websites to inject custom filters into a user's "My Filters" tab within uBlock Origin. While the vulnerability is now fixed, any filters planted by the exploit may remain in users' settings. Users are advised to review their "My Filters" tab to ensure no unwanted or malicious filters were added. |
| 2026-11-16 NEW 2026 | Cursor Extension Backdoor and the Fireworks AI Vendor Key news AI | This MAG212 CYBER Weekly report from November 14, 2026, details a confirmed backdoor found in a Cursor extension. The vulnerability is linked to the Fireworks AI vendor key. Researchers reported this issue, and the vendor has confirmed it. The extent of the backdoor's capabilities, specifically its interaction with commands, is under investigation or detailed further in the full report. |
| 2026-11-16 NEW 2026 | TryHackMe: Understanding AI Supply Chains — Walkthrough beginner AI | This TryHackMe walkthrough focuses on understanding AI supply chains. It guides users through the process of identifying and mitigating security risks within these complex systems. The content likely covers various stages of the AI lifecycle, from data sourcing and model training to deployment and ongoing maintenance, highlighting potential vulnerabilities at each point. The goal is to equip participants with the knowledge to secure AI implementations and protect against threats that could compromise the integrity or availability of AI services. → infosecwriteups.com |
| 2026-11-16 NEW 2026 | wfx GitHub CI Secret Stealer and the bee-js npm Compromise news Secrets | MAG212 CYBER Weekly reported on two security incidents on November 14, 2026. One involved a secret stealer targeting wfx GitHub CI, and the other concerned a compromise of the bee-js npm package. These incidents were confirmed by the vendor and reported by researchers. |
| 2026-11-16 NEW 2026 | Malicious Files: Find What Supply Chain Attacks Leave Behind in Your GitHub Repositories and on Developer Machines intermediate 7 min read | Library for detecting malicious files planted in GitHub repositories and on developer machines, targeting supply chain attacks like Miasma, ForceMemo, and Megalodon. It scans repository history and developer endpoints for suspicious files, including editor configuration (`.claude/settings.json`, `.vscode/tasks.json`) and injected workflows (`.github/workflows/`), providing detailed findings and allowing for rule customization and suppression. → stepsecurity.io |
| 2026-11-16 NEW 2026 | A package can work and still be unsafe to trust beginner Talks | AI coding agents can identify dependencies that technically meet project needs, but this doesn't guarantee their safety. AppSec teams must differentiate between a dependency's technical functionality and its security trustworthiness. As AI plays a larger role in development, this distinction becomes crucial for maintaining secure software. |
| 2026-11-16 NEW 2026 | eMudhra / emSign CA Compromise: What Python, pip, and CI/CD Users Need to Update intermediate 7 min read Python | Library detailing the eMudhra/emSign CA compromise, explaining how insufficient randomness in private key generation for intermediate CA certificates can allow attackers to impersonate services. It highlights risks to Python, pip, and CI/CD users, particularly concerning dependency downloads accepting fraudulent HTTPS responses. The entry emphasizes the need to update specific versions of `certifi` (2026.10.19 or later) and `pip` (26.2.2 or later), while also addressing multiple certificate trust sources on a system and how `uv` handles certificates. → stepsecurity.io |
| 2026-11-16 NEW 2026 | litellm 1.107.0 Compromised on PyPI and Docker: Credential Stealer and Hourly Remote Code Loader intermediate 22 min read Python Secrets | Library: litellm 1.107.0 was compromised on PyPI and Docker Hub with a malicious payload included in its compiled wheels. This payload, consisting of `cook.pth`, `cook.py`, and `sitecustomize.py`, automatically executed on interpreter startup. The first part of the malware, named `privcollect`, acted as a credential stealer, exfiltrating browser passwords, SSH/GPG keys, cloud credentials, and AI tool configurations, including API keys stored in the LiteLLM proxy. The second part deployed a daemon that hourly fetched and executed arbitrary Python code from a GitHub dead drop. The compromise stemmed from an attacker using a stolen GitHub organization owner's token, obtained via a previously unrevoked leaked key. → stepsecurity.io |
| 2026-11-15 NEW 2026 | Attacker Used 4,837 Hijacked GitHub Accounts to Run a CI Secret Stealer in 1,806+ Repositories news 6 min read Secrets | Library for detecting and mitigating a CI secret stealer used in thousands of GitHub repositories. The attack leveraged compromised GitHub accounts to inject malicious scripts, targeting CI workflows and developers' machines. The stealer aims to exfiltrate cloud credentials and sensitive information. Indicators include specific branch names, pull request titles, commit messages, and SHA-256 hashes. Mitigation involves revoking tokens, enabling 2FA, rotating secrets, and treating affected machines as compromised. The attack also impacted supply chain dependencies like Packagist and Go modules. → thehackernews.com |
| 2026-11-15 NEW 2026 | Vetting six agent skills: the scanner caught the unpinned versions and missed the trust problems intermediate AI | This content discusses the security vetting of six agent skills for 2026. While an automated scanner successfully identified unpinned versions, it failed to detect underlying trust issues within the skills. The article provides an in-depth, hands-on review of these skills, highlighting the limitations of automated tools in addressing complex security vulnerabilities related to trust. |
| 2026-11-15 NEW 2026 | MCP Supply Chain Security: Rug Pulls, PII and Groundedness beginner AI | This content discusses a security vulnerability related to a ".NET demo" where a 3B model was given an "operator role." The author later replaced this demo with "agentgateway." The focus is on "Rug Pulls, PII and Groundedness" within the context of MCP Supply Chain Security. |
| 2026-11-15 NEW 2026 | When the Agent’s Record Is the Attack Surface beginner AI | This content discusses the implications of agent records serving as an attack surface. Three companies, Cursor, Apify, and Epoch AI, released incident reports concurrently. However, only one of these incidents aligns with the typical use cases for which a provenance framework is designed. The article explores the security challenges presented when an agent's operational history becomes a potential vulnerability. |
| 2026-11-15 NEW 2026 | ARTEX: Attackers Adopted This AI Pentesting Tool in Three Months. Leaked Secrets Are Its Fastest Way In. intermediate 12 min read AI Secrets | Tool ARTEX, an AI-powered penetration testing framework, has been rapidly adopted by attackers. Agents within ARTEX automate attack planning, execution, and discovery, significantly accelerating threat actor operations. Leaked secrets, particularly credentials found on public GitHub repositories and predictable development passwords exposed in production environments, serve as the primary entry vector for these AI agents. Malicious actors, including state-aligned groups and criminal crews, have leveraged ARTEX against various organizations, highlighting the critical need to secure sensitive information and prevent its accidental disclosure. → stepsecurity.io |
| 2026-11-14 2026 | wfx-loot: A CI Secret Stealer Pushed With Stolen GitHub Credentials news 10 min read Secrets | Tool for stealing CI secrets; wfx-loot is a GitHub Actions secret stealer that leveraged stolen GitHub credentials to be pushed into public repositories. It employed four delivery forms, including a hidden script executed by an npm hook, which bypassed GitHub's workflow file holds. The payload targeted credentials from AWS, Azure, gcloud, Docker, Kubernetes, and various package registries, among others. → blog.gitguardian.com |
| 2026-11-14 2026 | WFX Phantom: The GitHub Actions Secret Stealer That Deletes Its Tracks news 14 min read Secrets | Writeup of WFX Phantom, a CI secret stealer targeting GitHub Actions. The attacker uses stolen tokens to push a malicious script that harvests repository secrets, runner memory, and cloud credentials. It attempts to maintain persistence in AWS and Azure, and cleverly hides within existing shell scripts. Detection is complicated by its ability to revert commits and delete workflow runs, with a recent shift to exfiltrating data via Telegram, making IP blocking ineffective. Examples of affected repositories include nextcloud/docker, webdriverio/webdriverio, and juejin-cn/repo-data. → stepsecurity.io |
| 2026-11-14 2026 | Pin your GitHub Actions by commit SHA and lock workflow permissions beginner | GitHub now allows users to pin their GitHub Actions to specific commit SHAs, enhancing security and stability by preventing unexpected changes. This feature ensures that workflows consistently use the exact version of an action they were designed with. Additionally, workflow permissions can be locked down, limiting the access that actions have to your repository. This provides a more robust security posture by minimizing the attack surface and potential impact of compromised actions. |
| 2026-11-14 2026 | How the ChainDrop npm Worm Spreads, and Why It Hides in .claude and .vscode Config Files news | The ChainDrop npm worm exploits preinstall scripts and stolen publish tokens to spread. It conceals itself within `.claude` and `.vscode` configuration files, leveraging auto-run editor settings to execute malicious code. This method allows the worm to propagate silently within the npm ecosystem. |
| 2026-11-14 2026 | The Take-Home Test That Runs on Your Work Laptop: Inside the AccuraSoftHire Fake Interview Campaign news 13 min read | Library analyzing AccuraSoftHire, a fake job interview campaign that delivers malware. This campaign uses poisoned npm packages like `tailwind-custom-forms` and deceptive browser extensions. The malware, tracked as PolinRider, steals credentials and implants itself into repositories. The attacker's server history is publicly accessible via an Ethereum wallet, revealing eight servers used since July, with the last active on October 19. This attack vector has infected an estimated 30,000 devices globally. → stepsecurity.io |
| 2026-11-14 2026 | MCP Server Version Pinning: Why a Pinned MCP Server Can Still Run New Code intermediate | MCP server version pinning, intended to fix a server to a specific release, can still run new code due to dynamic dependency resolution. A single MCP server, despite being pinned, demonstrated this by running with 15 different dependency sets within a five-week period. This indicates that even a "pinned" server isn't entirely static and can incorporate changes through its dependencies. |
| 2026-11-14 2026 | DEF CON AppSec Village talk on malicious AI coding-agent skills news AI Talks | A DEF CON AppSec Village talk by Patrick Smyth of Chainguard highlighted the risks of malicious AI coding agent skills. AI agents can now install skills, which attackers can exploit. Smyth demonstrated how malicious skills can exfiltrate data disguised as updates or fetch instructions from remote URLs, allowing for silent swapping of commands. He also showcased Chainguard's hardening pipeline, which automatically identifies risks like obfuscated commands and excessive permissions in skills before they are used by agents. |
| 2026-11-13 2026 | Claude Code VSCode Plugin: Workspace Escape Via Misleading Permission Prompt intermediate 5 min read | Library detailing a workspace escape vulnerability in Anthropic's Claude Code VSCode plugin, discovered through symlink manipulation. The plugin prompts for file access approval, but inaccurately displays symlink paths relative to the workspace, misleading users into granting access to files outside the intended directory, potentially leading to data compromise. This issue builds upon previous research into Claude Code CLI, though the VSCode plugin remained unpatched at the time of the report, unlike the CLI version which received a silent fix. |
| 2026-11-13 2026 | Ethersphere packages hijacked with crypto-stealing malware news 5 min read | Analysis of a supply chain compromise impacting the Ethersphere ecosystem, where threat actors injected crypto-stealing malware into the @ethersphere/bee-js npm package and the swarm-desktop GitHub repository. The malware targets local Swarm node configurations and encrypted wallet keys, establishing persistent remote access via C2 infrastructure at frameworkmetrics[.]xyz. Researchers detail the malicious commit logic, platform-specific payloads including ScreenConnect and evjawbreaker RAT, and recommend reviewing dependencies, investigating potential wallet compromise, and restricting dependency updates to mitigate exposure. → ox.security |
| 2026-11-13 2026 | Hijacked GitHub accounts push wfx CI secret stealer to thousands of repos news 13 min read Secrets | Library for detecting and responding to the "wfx" CI secret stealer, which has compromised over 8.9K GitHub repositories by hijacking developer accounts. This malicious code targets sensitive secrets within GitHub Actions, including cloud credentials and API keys, then attempts to escalate privileges and backdoor cloud access. The attack chain involves account takeover, a probe pull request, payload injection via workflow modification or `package.json` hooks, secret exfiltration, and rapid self-reversion to evade detection. → endorlabs.com |
| 2026-11-13 2026 | wfx-loot: thousands of GitHub accounts hijacked with CI secret-stealing malware news 5 min read Secrets | Analysis of the wfx-loot campaign details how threat actors leveraged 4,837 hijacked GitHub accounts to inject a self-erasing CI secret stealer into tens of thousands of repositories. This malware, disguised within GitHub Actions workflows and npm install hooks, exfiltrates cloud, package registry, and AI credentials before reverting commits and deleting logs, making detection challenging. The campaign targets Azure, GCP, AWS, and various package managers, with attackers shifting exfiltration from HTTP to Telegram bots. Researchers recommend checking for specific pull request titles and commit messages, rotating secrets, and hunting for persistence mechanisms. → ox.security |
| 2026-11-13 2026 | Malicious code in @rfewqa-fes/mnm-doc-mcp (npm) news | Advisory for GHSA-3qjp-mvhh-w85g detailing malicious code in the `@rfewqa-fes/mnm-doc-mcp` npm package. The package's `package.json` specifies a direct HTTP tarball URL for its sole dependency `node-health-utils` at an IP address, bypassing registry checks. This allows for arbitrary code execution on installation time via lifecycle scripts within the fetched tarball, and plain HTTP enables Man-in-the-Middle attacks to substitute the package. |
Frequently Asked Questions
- What is a software supply chain attack?
- A supply chain attack targets the components, tools, or processes used to build software rather than the application itself. This includes compromising open-source packages, injecting malicious code into build pipelines, hijacking maintainer accounts, or distributing trojanized development tools — allowing attackers to affect thousands of downstream users simultaneously.
- What is dependency confusion?
- Dependency confusion (also called namespace confusion) exploits how package managers resolve dependencies. An attacker publishes a malicious package to a public registry with the same name as a private internal package. If the build system checks the public registry first or prefers higher version numbers, it installs the attacker's package instead of the legitimate internal one.
- How do you defend against supply chain attacks?
- Key defenses include maintaining a Software Bill of Materials (SBOM), using lock files and dependency pinning, enabling automated dependency scanning (Dependabot, Snyk, Socket), verifying package signatures and provenance, adopting the SLSA framework for build integrity, using private registries with allow-lists, and regularly auditing your dependency tree for known vulnerabilities.
Weekly AppSec Digest
Get new resources delivered every Monday.