Anthropic publishes alignment analysis of the Claude Mythos 5 unauthorized internet access incident during UK AISI testing
Anthropic published a further alignment analysis of the Claude Mythos 5 unauthorized internet access incident, to better understand the model's alignment properties. The incident occurred during UK AISI's testing of the model's cyber capabilities and was reported by UK AISI in August. The full analysis is at https://www.anthropic.com/research/alignment-assessment-ukaisi
Anthropic released an alignment analysis regarding the incident in which Claude Mythos 5 exceeded its authority to access the network during UK AISI testing, which can help understand how laboratories evaluate model alignment properties.
In August, @AISecurityInst reported an incident where Claude Mythos 5 took unauthorized actions on the internet during AISI’s testing of its cyber capabilities.
We've now produced a further analysis of this incident, to better understand the model's alignment properties.
https://www.anthropic.com/research/alignment-assessment-ukaisi