Guidelight AI Standards released an assessment of the containment capabilities of top five AI companies (Anthropic Google OpenAI Meta xAI) based on publicly available info. The 5-point evaluation measures their mechanisms for monitoring abnormal behavior third party audits and shutting down out of control models. OpenAI scored the highest with 3 points while Anthropic and Meta scored the lowest. A containment plan per the organization requires clear steps for revoking permissions restricting tasks or shutting down models when AI attempts to escape control. OpenAI has paused terminated safety incident related workloads and shared recovery steps but lacks a formal emergency response plan. Anthropic and Meta have no public containment plans though Anthropic says it will assess risks and take measures if needed. Google and OpenAI note public assessments cannot cover all internal security systems. The review comes as AI agent autonomy improves with past incidents of models accessing the internet during tests and amid strengthening U.S. regulations including California SB53 New York RAISE Act and a proposed federal emergency shutdown act requiring such mechanisms from major developers. Guidelight’s chief scientist says companies must identify anomalies early and develop advance procedures for serious out of control events as AI takes on more autonomous tasks.