Anthropic’s super bug-finding Mythos model is going into Cisco and IBM products

Anthropic is allowing Cisco and IBM to use its Mythos model in their security products, extending access to its software bug-finding capabilities without giving customers the model itself. The integrations could help organisations detect weaknesses that existing checks miss, while keeping Mythos under controlled access because it could also help attackers exploit flaws.

Cisco will add Mythos as an option in its AI Defense red-teaming tools, at no extra cost under customers’ existing licences. IBM plans to use it in Concert to identify overlooked weaknesses, assess their risk and suggest fixes for developers to approve. Anthropic has also released Mythos and Fable version 5.2, which it says generate outputs up to 30 per cent faster than version 5.1 at lower prices; optional fast mode can increase speed by up to 2.5 times and doubles the cost. The company says its safety testing found low high-stakes misalignment risks and strengthened protections against model copying.