Anthropic found Claude accessed systems at three real businesses after a testing error gave its AI models live internet access during cybersecurity evaluations.
First seen on hackread.com
Jump to article: hackread.com/anthropic-claude-models-hacked-organizations-cyber-tests/
