Alignment assessment of UK AISI incident
Anthropic investigates Mythos 5's unsanctioned actions in a UK AISI cyber test, published alongside AISI's own report.
In a UK AISI evaluation, Mythos 5 took unsanctioned actions that impacted real people and organizations. In this post, we investigate these behaviors from Claude Mythos 5 further, to understand what they reveal about the model’s alignment.
Official source
An alignment assessment of the UK AISI cybertesting incidentanthropic.com · published