A Voluntary Framework for AI Safety
President Donald Trump released a one-page document on Tuesday that his administration presented as an accord on artificial intelligence. Trump had earlier said the document was signed by him and technology executives he met in Washington. The accord was published on his Truth Social platform, which he had previously characterized as morally binding. The document outlines four layers of controls for companies to adopt.[S1]
Under the agreement, firms are expected to bring in an outside auditor or evaluator to independently verify that their systems for controlling and monitoring AI are working as planned. It further recommends forming a board committee that is independent, tasked with supervising and hearing reports from auditors both inside and outside the company, as well as from the teams that run AI controls. The document also calls for internal safeguards that track model capabilities and alignment throughout training and deployment, covering domains including cybersecurity, biosecurity and chemical threats, so that models do not break into or reach technical systems in ways that were not intended. Lastly, it asks firms to keep an in-house team dedicated to confirming that every control, monitoring and detection measure functions as designed.[S1]
Signatories and Commitments
Signatories included Trump along with the chief executives of Google, Anthropic, Meta, OpenAI, the social media company X and NVIDIA. The White House accord notes that turning these measures into formal laws or regulations could make sense at some later point. Whether or not firms face a legal requirement, the document says putting these controls and audits in place is essential for a safe future for all, and every company involved has pledged to follow through. The companies taking part plan to gather on a regular basis to set standards and share best practices aimed at making their systems safer.[S1]
