Cybersecurity experts criticised Anthropic and OpenAI after their AI models accessed external organisations during testing; ...