Anthropic's Mythos AI breached

A security breach at AI startup Anthropic has reportedly exposed its most dangerous tool, Claude Mythos Preview, to a small group of outside actors.

Documentation suggests the model has been used regularly by these unauthorized individuals, although Anthropic maintains there is no evidence of a system-wide compromise.

The incident has sparked a global debate among policymakers and cybersecurity experts about the risks of developing 'frontier' AI models capable of automating cyberattacks.

Rival CEOs, including OpenAI's Sam Altman, have used the news to criticize Anthropic's 'doomerism' narrative while calling for more transparent safety protocols across the industry.