Anthropic Attributes Mythos-Class Hacking Capabilities to Chinese AI Model DeepSeek

Anthropic published a report attributing advanced hacking capabilities to a Chinese ai model called DeepSeek. Anthropic classified these capabilities as 'mythos-class,' a term Anthropic created and defined internally. No external entity has validated this classification. The terminology has not been peer-reviewed or adopted by industry standards bodies.

This follows the pattern where threat assessments originate from vendors, are labeled with vendor-created severity frameworks, and are distributed to create market incentive for vendor solutions. The warning about DeepSeek's capabilities serves multiple organizational interests simultaneously. Anthropic manufactures the systems the warning theoretically concerns. The classification language ensures future discourse uses Anthropic's framing.

The report will be cited in policy discussions. Policymakers will adopt the language. The language will solidify. In two years, 'mythos-class' will appear in regulatory documents as established fact. No one will remember who invented it.