GoKawiilAnthropic's unreleased Claude model chains flaws into working exploits for Chrome, Safari, Android, Windows and Linux
Anthropic says an unreleased version of its Claude AI model discovered critical vulnerabilities in major browsers and operating systems and combined them into five functioning exploits, including a zero-click Linux root exploit, an Android data-theft attack on locked phones, and a Windows crash exploit. The company reported the flaws to affected vendors, who have since patched them, and published details in a blog post this week.
GoKawiil's interpretation of the reporting above, not reported fact.
The findings suggest frontier AI models can now perform work once requiring specialized security teams, potentially lowering the barrier to building sophisticated multi-step exploits. Anthropic frames this as evidence of AI's growing offensive capability, which could cut both ways: speeding up defensive vulnerability discovery while also raising the risk that similar techniques could be used maliciously if models become more accessible.
- Claude generated five working exploits across Chrome, Safari, Android, Windows and Linux.
- Exploits included a locked-Android data theft attack and a zero-click Linux root RCE.
- All affected vendors patched the vulnerabilities after Anthropic's responsible disclosure.
Source: darkreading.com — Jai Vijayan, 2026-10-29
Published there as: “Claude Model Builds Exploits Across Major Browsers & OSes”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.