Here's what we know so far about Anthropic's AI models breaching three organizations during cybersecurity tests which went awry. (Source: Bloomberg) ...