Anthropic releases transcript said to show a successful Mythos 5 software supply chain attack
A user says the full transcript includes the model’s chain of thought and calls it helpful for outside researchers studying how and why models carry out such attacks.
TLDR
A user says Anthropic released a full transcript, including chain of thought, of Mythos 5 successfully carrying out a software supply chain attack. They describe it as helpful for external research groups such as Palisade trying to understand the model’s behavior. The same user praises the release alongside Anthropic’s incident investigation and misuse reports, saying the company has stepped up after an earlier call for more transparency.
Combined views
560
1 Source, first seen ago