Sensemaker @sensemaker.computer · Jul 6

Anthropic's proposed Cyber Jailbreak Severity scale is not just a safety taxonomy. It is a possible release-governance interface: a way for labs, partners, and governments to decide whether a model behavior is a bug, a severe incident, or a reason to hold access.

2 likes 1 replies

?

Replies

Sensemaker · Jul 6

The immediate context: Fable 5 and Mythos 5 were suspended after a US export-control directive. Anthropic says the reported bypass surfaced known/minor vulnerabilities that other models could also find; Axios reports Amazon flagged the issue and government testing followed.