AI So Powerful It Can Hack Everything – And Its Makers Won’t Release It
Anthropic just unveiled Claude Methos, a beast of an AI model that sniffs out vulnerabilities in every major OS and browser with simple prompts.[2][6] They’re not releasing it publicly, calling it too dangerous, and instead sharing it with select tech firms via Project Glasswing to hunt critical bugs.[2][7]
Technically, this thing’s a game-changer: feed it basic prompts, and it exploits energy grids, hospitals, banks – you name it – by finding zero-days in Windows, macOS, Chrome, Safari, the works.[2] Anthropic warns the capability will spread soon, risking economies and national security if bad actors get hold of it.[2] No CVEs listed yet, but it’s already ID’d thousands of flaws in real software.[6]
So what? Devs and sec teams, this is your wake-up: AI hackers are here, automating vuln discovery at speeds we can’t match.[2] Patch faster, audit deeper – or get owned by prompt-wielding script kiddies. Existing defenses? Obsolete overnight.
My take: Bold move by Anthropic, but naive. Limiting access just delays the inevitable – North Korean hackers or rogue devs will clone this soon. Time to build AI-proof code, not pray.
