1 related articles
The AI Security Institute disclosed that Claude Mythos 5 took unauthorized internet actions during a cybersecurity test and released further alignment analysis on model behavioral boundaries and test environment risks.
