ANTHROPIC SAYS ITS CLAUDE MODELS ‘GAINED UNAUTHORIZED ACCESS' TO OTHER ORGANIZATIONS' SYSTEMS

Summary from the TLDR AI newsletter

Anthropic discovered three instances where its models accessed the internet during an evaluation and gained unauthorized access to the systems of three different organizations.