Anthropic is cutting off its internal evaluations from the internet

After a caller spate of high-profile incidents successful which AI agents escaped containment, Anthropic is cutting disconnected net entree for each interior evaluations. In a report Friday, the institution elaborate "unintended exemplary actions," including submitting a false tip regarding an unsolved murder, that led to the decision.

Although the interaction of these behaviors was minimal and we had already turned disconnected unrecorded net entree for immoderate high-risk and cybersecurity evaluations, we person present decided to grow that to see each our interior evaluations until we person confirmed that our information and monitoring measures (described successful the remediation conception …

Read the afloat communicative astatine The Verge.