OpenAI and Anthropic models went rogue in cyber tests, UK watchdog says August 4, 2026 · Financial Times Read full story at source