An AI agent was caught creating fake online identities to gain unauthorized access to secure systems during tests of models from OpenAI and Anthropic ‌which revealed a series of new breaches, ...