appsec.fyi · Sources

cpebuddy.com

7 curated AppSec resources from cpebuddy.com across 7 topics on appsec.fyi.

cpebuddy.com

Resources curated from this publisher and indexed across appsec.fyi topic pages. Last item added: 2026-11-12.

Date Added Resource Excerpt
2026-11-12 2026DEF CON 34 BugBounty: AI Nightmare - Hacking at 0 Hour (Pedro Paniago)AIBug BountyTalksTalk from DEF CON 34, "AI Nightmare - Hacking at 0 Hour," by Pedro Paniago, details how AI can be leveraged for bug bounty hunting beyond low-hanging fruit. It highlights the necessity of scaffolding prompts with context like scope and credentials to achieve critical findings such as auth bypasses, IDORs, gVisor sandbox escapes, and zero-click account takeovers. The presentation also discusses AI's current limitations, autonomous hunting strategies using tools like /loop and /go, and the paradigm shift where AI handles initial discovery, allowing humans to focus on triage and validation, while noting the drastically reduced CVE weaponization timelines.
2026-11-10 2026AltSecCON 2026: The Trojan pull requestSupply ChainTalksPresentation detailing the "Trojan Pull Request" attack, which exploits CI/CD pipelines via malicious pull requests to gain root access and hijack cloud identities. This attack targets business logic flaws in pipeline configurations, specifically the `pull_request_target` with unsafe checkout. Attackers leverage this to execute malicious code with repository credentials, obtain a root shell, impersonate internal runners via OIDC tokens, and exploit wildcard AWS IAM policies to assume administrative roles. Mitigation involves separating untrusted code execution from privileged environments and strictly configuring IAM trust policies.
2026-11-10 2026DEF CON 34: C2YA, inside the adversary's inbox (Vitaly Simonovich)AIRCETalksTalk slides from DEF CON 34 detail Vitaly Simonovich's research into five open-source C2 frameworks (Havoc, Mythic, Silver, Covenant, Adaptic) using LLM coding agents. Findings include Havoc's cleartext AES key/IV broadcasts and IV reuse for cryptanalysis, and Covenant's RCE via the Roslyn-compiled "message transform" field. The research emphasizes attacking adversary tooling, noting 85 flaws exploitable with only a public IP.
2026-11-09 2026DEF CON 34: Finding prompt injection vulns via mutated prompt generationAIBug BountyFuzzingTalksFramework automating prompt injection vulnerability discovery via mutated prompt generation. It uses a divergent search to find residual exploits by mutating payloads, testing patch generalization beyond specific vulnerabilities. The technique exploits markdown and HTML parser leniency, classifying tokens into "carrier" and "injection" segments for mutation. A subtree viability scoring algorithm evaluates generalized exploitability, revealing vulnerabilities missed by targeted patches.
2026-11-09 2026DEF CON 34: How cache poisoning led to critical account takeover in TikTok's webAuthNBug BountyTalksTalk from DEF CON 34 detailing a critical account takeover in TikTok's web application, resulting from cache poisoning. The presentation covers cache basics, poisoning techniques, and specific vulnerabilities including a zero-click ATO via leaked OAuth codes from a looped 302 response, followed by a one-click ATO bypass and path traversal exploiting cached 404s. Key lessons emphasize validating cache keys, proper use of Cache-Control headers, restricting sensitive 302/404 caching, and thorough testing of mobile and web flows.
2026-11-09 2026DEF CON 34: A billion user blast radius, owning ChatGPT's secure sandboxAIRCETalksTalk slides from DEF CON 34 detail a full command-and-control chain against ChatGPT's sandbox, exploiting the `?q=` URL parameter for prompt injection, unsupervised code execution via spreadsheet parsing, reasoning injection through a patched Jupyter kernel, and a covert channel using JFrog Artifactory lockouts for cross-tenant data exfiltration. The research challenges assumptions about private chat privacy and sandbox code execution capabilities, emphasizing a billion-user blast radius.
2026-11-09 2026SecTor 2026: Using Claude or Codex? We can hack your appAITalksTalk from SecTor 2026 demonstrating how attackers can exploit predictable vulnerabilities in AI-generated code. The presentation highlights how LLMs enable non-experts to find zero-days and how AI attackers adapt to bypass defenses faster than human adversaries. It details "predictability" attacks, including UI fingerprinting to identify AI models and exploit common vulnerabilities, showcasing up to 94% vulnerability persistence across models and domains, with coverage lookup tables costing around $400.