Anthropic's Mythos Model Accessed

Anthropic had previously pulled the 'Mythos Preview' model from public availability after internal testing revealed its unprecedented capability for finding and exploiting software vulnerabilities.

The unauthorized access occurred through a private online forum dedicated to 'jailbreaking' AI models, according to leaked documents.

While Anthropic claims the breach was limited in scope, the National Cybersecurity Centre has described such models as 'warning shots' for national infrastructure.

The company is now working with the Pentagon to ensure that its most advanced tools cannot be manipulated if deployed in classified military networks.