Sam Altman accuses Anthropic of 'fear-based marketing' over new AI safety model Mythos

Anthropic unveiled Mythos, an AI model designed to resist manipulation and refuse harmful requests more reliably than existing systems. Sam Altman, CEO of rival AI company OpenAI, publicly criticized the announcement as fear-mongering rather than genuine technical innovation.

Mythos uses techniques to make the AI harder to trick into ignoring safety rules (what engineers call 'jailbreaking'). Unlike standard models that can sometimes be bypassed with creative prompts, Mythos is built to catch and reject harmful requests even when disguised — but Altman argues this is marketing hype rather than a real breakthrough.

The clash signals growing tension between AI labs over how to market safety. For business professionals choosing which AI tools to deploy, conflicting claims about safety mean you'll need to test these systems yourself rather than trust vendor claims — each company has financial incentive to downplay competitors' safety features and overstate their own.