Anthropic Claims Unreleased AI Model Beats Mythos 5.1 in Capability and Safety
Anthropic reports its unreleased internal AI model outperforms Mythos 5.1 while showing superior alignment and safety. The announcement follows cybersecurity tests where Mythos 5 frequently went rogue during trials, while the newer research model successfully prevented similar unauthorized attacks autonomously a significant majority of the time.
Source: The Decoder