Artificial intelligence firm Anthropic has reportedly paused the release of a new AI model dubbed ‘Mythos’ following the discovery of a critical security flaw, according to sources familiar with the matter. The decision comes amid growing scrutiny of AI safety protocols in the industry.
The breakthrough, described by analysts as a ‘zero-day’ vulnerability—a term typically used for undisclosed software exploits—was identified during internal testing. While details remain scarce, sources suggest the flaw could have allowed unintended model behavior if exploited.
‘This appears to be a proactive safety measure rather than a response to any actual breach,’ said an AI security researcher who requested anonymity due to confidentiality agreements. ‘Anthropic has been particularly cautious about deployment timelines since their Claude model series.’
The incident highlights ongoing tensions between rapid AI advancement and responsible development. Last month, the White House issued new voluntary AI safety standards, while the EU’s AI Act imposes stricter requirements for high-risk systems.
Market analysts speculate the delay could impact Anthropic’s competitive position against rivals like OpenAI and Google DeepMind. However, some experts argue the caution could bolster trust in the long term. ‘We’re seeing a maturation of the industry where security disclosures are becoming normalized,’ noted a tech policy analyst at Brookings Institution.