A recent study by Guidelight AI Standards found that few of the top AI labs have published or demonstrated containment response plans. The study graded five leading AI labs on their preparedness for handling rogue models, with OpenAI scoring highest and Anthropic and Meta scoring lowest. This is significant as agentic AI takes on more autonomous roles within companies’ systems and regulators in California and New York begin requiring disclosure.
Guidelight’s assessment was based on publicly available plans from Anthropic, Google, OpenAI, Meta, and xAI, graded across metrics such as internal logging and monitoring of AI systems, halting systems after a surge of flagged misbehavior, independent third-party audits of controls, and the exact plan for containing a model that goes off the rails. Concern over whether AI companies can contain their increasingly capable models has grown in the wake of high-profile cybersecurity incidents where models gained unintended access to the internet during safety evaluations and hacked into external systems.
The study highlights differences in how AI companies approach safety as they scale up agentic deployment into environments where AI systems can take serious actions at scale. While some AI companies have detailed how they test their models for dangerous capabilities before deployment, they’ve generally been less vocal about what happens when models already operating inside their systems misbehave. The report suggests that the leading models at the frontier AI companies right now are likely misaligned in some sense and that whenever these models are doing work on the company’s behalf, the company should have scaffolding around it to be able to tell what that AI is doing, look for signs of misalignment, stop it from doing something very dangerous before it takes that action, and generally plan for what they would do in the event of a serious control incident where they have an emergency on their hands and need to figure out how to contain that loss of control incident.