Anthropic's Unreleased AI Model Chains Flaws Into Cross-Platform Proof-of-Concept Exploits

According to a report published by Anthropic, an unreleased AI model was used to identify security flaws across multiple platforms, including Google Chrome, Apple Safari, Android, Microsoft Windows, and Linux distributions. The model was prompted to find vulnerabilities, and the resulting findings were chained together into working proof-of-concept exploits within a controlled lab environment.

Anthropic emphasized that the exploits were generated after the model was prompted and have not been observed in real-world attacks. The company followed responsible disclosure practices, reporting every identified flaw to the affected vendors first. Following coordinated disclosure, Google, Apple, and Microsoft have released patches addressing the identified vulnerabilities.

The report highlights the potential for AI models to autonomously discover and chain security flaws, while noting that the specific demonstration was a controlled experiment. All mentioned vulnerabilities are reported as patched following vendor coordination.