Mythos Preview can weaponize N-day vulnerabilities in hours

Anthropic's research demonstrates that advanced language models (Mythos Preview) can rapidly turn disclosed, patched vulnerabilities into working exploits — producing multiple Firefox and Windows kernel proofs-of-concept and several privilege-escalation chains within minutes to hours and at modest cost, which shortens exploit development timelines from days/weeks to hours and increases risk from N-day to "N-hour."