Anthropic's 'Mythos' AI Raises Major Cybersecurity Alarms

Anthropic has unveiled Claude Mythos Preview, its most advanced AI model yet, named after the Greek word for story or narrative. The system scored 97.6% on the US Mathematical Olympiad and 93.9% on SWE-bench, marking a significant leap over previous models. The model raised serious cybersecurity concerns after independently discovering thousands of unknown software vulnerabilities, including a 27-year-old OpenBSD flaw. Anthropic has restricted public access, launching Project Glasswing, a limited initiative with partners including Amazon, Apple, Google, and Microsoft.