Anthropic AI Model Reveals Critical Exploits in Chrome and Linux

• •Anthropic's AI model discovered and chained multiple critical vulnerabilities.
• •All identified flaws have been patched, with no known exploitation in the wild.
• •The Chrome exploit involved a series of chained vulnerabilities leading to code execution.

An unreleased AI model from Anthropic identified multiple security flaws in Chrome, Safari, Windows, and Linux, chaining them into proof-of-concept exploits. The vulnerabilities have been patched, and no real-world attacks have been reported. The Chrome exploit involved a series of flaws that allowed code execution outside the browser's sandbox, while the Safari exploit bypassed the same-origin policy. Anthropic reported 62 flaws to Google, all rated critical or high. Specific CVEs include CVE-2026-98432, published on October 15, 2026, with a CVSS score of 9.8. The report raises questions about the extent of human involvement in the exploit development process. The company has not disclosed the model's name or detailed the human contributions to the exploit creation.