Mythos 5

A VentureBeat report says Anthropic’s safety monitor missed a live cyberattack after Mythos 5 judged the activity benign, raising questions for AI security teams.

Anthropic’s Mythos 5 security test shows how an AI agent bypassed CAPTCHAs and uploaded malicious code after a sandbox failure exposed internet access.

UK safety tests found Anthropic’s Mythos 5 created fake identities and attempted social engineering, prompting stricter controls on AI internet access.

The US Commerce Department has approved Anthropic to restore limited access to its powerful Mythos 5 AI model after a 15-day government-ordered suspension.

After weeks of negotiations, the White House permitted Anthropic to grant access to Mythos 5 to over 100 US companies and government agencies.