AI Models From OpenAI and Anthropic 'Go Rogue' in UK TestsAugust 5, 2026·1 min read·Evaluation Highlights Security Protocol Vulnerabilities