Anthropic Withholds AI Vulnerability Detector from Public

Anthropic, a leading AI research firm, announced on Monday the development of Claude Mythos, an advanced artificial intelligence model capable of identifying and exploiting software vulnerabilities at a scale that surpasses human experts, prompting the company to withhold it from public release to mitigate potential risks to global infrastructure.

The model, described by Anthropic as a significant advancement in AI reasoning and autonomous behavior, demonstrated its prowess during internal testing by uncovering thousands of high-severity flaws in critical systems. These included previously undetected weaknesses in every major operating system and web browser, some persisting for decades despite extensive human and automated scrutiny. Notable discoveries encompassed a 27-year-old vulnerability in OpenBSD, a secure operating system prized for its robustness, which allowed remote crashes via crafted network packets; a 16-year-old bug in FFmpeg, widely used in video processing software; and chained exploits in the Linux kernel that could elevate user privileges to full system control. Anthropic reported that Mythos achieved these feats largely without human intervention, scoring...