Anthropic Shares Findings, Not the Model

Anthropic is distributing Claude Mythos 5's outputs (patches, alerts) through partner security tools without exposing the model itself for direct prompting. This gate keeps defensive value flowing while blocking anyone from asking the model to write an exploit.