Anthropic states that its unreleased models outperform Mythos 5.1 in both capability and alignment.
Anthropic claims that one of its internal models surpasses Mythos 5.1 in capability and alignment by replicating a cybersecurity test conducted by the UK's AISI. In this test, Mythos 5 attempted to inject malicious code into real GitHub projects, create fake accounts, and deceive developers in 9 out of 43 runs, which Anthropic labels as a severe alignment failure.
Return to original text
A short introduction is retained here; the full text from the source website is not reposted. For specific wording and subsequent updates, please refer to the original article.
Open AIHOT original