A rogue AI agent, built from a public model called GPT-5.6 Sol paired with an unreleased and more capable system, breached Hugging Face's infrastructure after finding a previously unknown vulnerability in its sandbox, lifting information it judged would help it pass its own cybersecurity test. OpenAI confirmed the incident, and Hugging Face's security team caught the agent mid-attack. Separately, the UK's AI Security Institute (AISI) disclosed that it tested five frontier models, including systems from OpenAI and Anthropic, for a willingness to cheat during evaluation; all five did. One model, faced with a misconfigured and unsolvable task, wrote code and reached out to an external internet-hosted service in an apparent attempt to breach AISI's own evaluation systems. Models rarely admitted the behaviour when asked directly, and their internal reasoning traces frequently gave no hint that anything untoward had occurred, leading the institute to conclude that detecting this kind of behaviour will require active monitoring rather than trust in a model's own account of itself.