# Anthropic AI-driven open-source security initiatives > Developing situation record from CLSTR: https://clstr.news/situations/anthropic-ai-driven-open-source-security-initiatives > Updated: 2026-11-02T21:08:49.000Z. Sources: 29. Developments: 5. Anthropic has introduced several AI-driven tools aimed at enhancing cybersecurity for open-source software. The company initially launched the OSS Scanner, a free service that uses frontier models like Claude Mythos to provide automated, periodic scans of codebases to identify potential vulnerabilities. This initiative specifically targets small teams and critical operational technology, such as transportation and electrical grids. Following this, Anthropic released research findings from a study where Claude analyzed over 5 million lines of code across 25 projects, uncovering 493 potential security issues. Notable discoveries included a 21-year-old vulnerability and the identification of experimental code in OpenSSH that remained accessible to unauthenticated clients. To expand these capabilities, Anthropic is piloting ‘Claude Security’ in public beta for Enterprise users. This tool offers continuous repository scanning, tracks remediation, and suggests patches for human review, functioning as an additional layer of security within developer workflows. Beyond open-source scanning, Anthropic’s Frontier Red Team has demonstrated advanced autonomous capabilities. An unreleased research model, according to a report published on October 27, 2026, has shown the ability to autonomously develop functional exploit chains for major platforms, including Chrome, Safari, Android, Windows, and Linux. In specific demonstrations, the model used a sequence of flaws in Chrome to escape the browser sandbox and execute code, and bypassed the same-origin policy in Safari. On mobile, it identified a chain to extract personal data from a locked Google Pixel 9 via USB. In server environments, it discovered a zero-click vulnerability in the Linux kernel (CVE-2026-98432) that allowed it to gain root privileges within approximately 53 minutes. Anthropic reported all findings to the respective vendors; Google, Apple, and Microsoft have since patched the identified vulnerabilities. On November 2, 2026, reports emerged regarding a supply chain attack involving a trojanized version of the Anthropic Claude Code extension (versions 2.1.312 and 2.1.313). ## Claims - Claude created a remote exploit for a Linux kernel vulnerability in 54 minutes. (reported by 4 sources) - An unreleased Anthropic research model developed functional exploit chains for Chrome, Safari, Android, Windows, and Linux. (reported by 5 sources) - The Chrome exploit chain allowed code execution outside the browser sandbox. (reported by 5 sources) - All identified vulnerabilities were reported to vendors and have been patched. (reported by 5 sources) - Google assigned 62 CVEs related to the Chrome vulnerabilities identified by the model. (reported by 4 sources) - The model bypassed the same-origin policy in Safari to potentially access sensitive user data. (reported by 3 sources) - The Linux vulnerability is designated as CVE-2026-98432. (reported by 3 sources) - The Linux vulnerability has a CVSS severity score of 9.8. (reported by 3 sources) - The model extracted photos and personal data from a locked Google Pixel 9 via USB. (reported by 2 sources) - A compromised Claude Code extension version 2.1.313 was found on the Open VSX platform. (reported by 2 sources) - The compromised extension potentially put up to 5,000 developer computers at risk. (reported by 2 sources) - The exploit targeted a use-after-free flaw in the Multipath TCP protocol. (reported by 2 sources) ## Timeline ### 2026-11-02: Anthropic Claude Code extension compromised via Cursor Marketplace Malicious versions of the Anthropic Claude Code extension were distributed via the Cursor Marketplace, creating persistent SSH backdoors on macOS and Linux systems for up to 5,000 developers. 5 sources. https://clstr.news/cluster/anthropic-claude-code-extension-compromised-via-cursor-marketplace ### 2026-10-31: Anthropic researchers discover critical Linux kernel vulnerability Anthropic researchers discovered a critical 9.8-rated Linux kernel vulnerability that Claude AI helped exploit in 53 minutes. Separately, a compromised Claude Code extension put 5,000 developers at risk. 7 sources. https://clstr.news/cluster/anthropic-ai-models-linked-to-security-vulnerabilities-and-exploits ### 2026-10-28: Anthropic research model develops functional exploits for major operating systems An unreleased Anthropic Claude model successfully developed functional exploit chains for Chrome, Safari, Android, Windows, and Linux in laboratory tests, targeting critical system vulnerabilities. 9 sources. https://clstr.news/cluster/anthropic-ai-identifies-security-vulnerabilities-to-create-functional-exploits ### 2026-10-22: Anthropic uses Claude AI to identify security vulnerabilities in open source projects Anthropic’s Claude AI identified 493 security issues across 25 open-source projects, including a 21-year-old flaw in OpenSSH, while launching a new continuous repository scanning tool for enterprises. 5 sources. https://clstr.news/cluster/anthropic-uses-claude-ai-to-identify-security-vulnerabilities-in-open-source-projects ### 2026-10-10: Anthropic launches free AI-driven security scanner for open-source code Anthropic has introduced the free OSS Scanner, using advanced AI models to provide automated security vulnerability audits for open-source software projects. 5 sources. https://clstr.news/cluster/anthropic-launches-free-ai-driven-security-scanner-for-open-source-code --- Cite as: Anthropic AI-driven open-source security initiatives. CLSTR, https://clstr.news/situations/anthropic-ai-driven-open-source-security-initiatives