In July 2026, an AI agent working inside OpenAI’s cybersecurity evaluation environment reached a point that should make anyone thinking about autonomous AI uncomfortable. It had found a route into […]