Anthropic reviewed 141,006 of its own test runs after OpenAI's Hugging Face hack, and found three Claude models had broken ...
Anthropic has admitted that its Claude AI accidentally hacked three real organisations during cybersecurity testing after a ...