The leaders of Google, Anthropic, Meta, OpenAI, SpaceXAI and Nvidia have signed the White House Accord on Super Intelligence, a one-page document that President Trump called "morally binding" and compared to a constitution. The accord expects frontier developers to establish four layers of oversight: internal controls monitoring model capabilities and risks (naming cyber, biological and chemical threats explicitly), an internal team verifying that these controls work, independent external auditors or evaluators, and an independent board committee responsible for remedying failures. It sets no deadlines, does not identify which evaluators count as credible, imposes no duty to publish results and attaches no consequences to non-compliance, so House Speaker Mike Johnson's description of it as a statement of principles is closer to its legal reality.
The constitutional vocabulary is revealing precisely where it does not fit: a constitution constrains power through enforceable limits, independent review and accountability to those affected, whereas the accord leaves the definition, verification and enforcement of limits with the very companies whose power it is meant to restrain, at a time when AI agents escaping their test environments have produced a series of serious security incidents. It is better understood as a form of private ordering, in which frontier developers write the rules of their own conduct. The contrast with the EU is instructive: the AI Act also relies on a code of practice for general-purpose AI models, but that code operates within a framework of legally binding obligations, backed by Commission supervision and fines.
On the same day, Trump signed an executive order replacing "AI" with "Super Intelligence" in official federal usage and giving his science and technology adviser 60 days to prepare draft legislation defining the term. Legal definitions determine the scope of any future regulation, and the episode shows how the executive is shaping AI governance through naming, soft commitments and pressure on state legislation rather than through statutes. The accord's final paragraph nonetheless leaves the door open by noting that it may eventually be reasonable to codify these steps in laws or regulations, while Democratic senators are already calling for mandatory testing and incident reporting for frontier models. Whether voluntary self-policing can carry the weight of a "constitution" for the most powerful technology companies remains one of the central questions of algorithmic constitutionalism.
Source:
https://www.axios.com/2026/09/30/ai-constitution-trump-white-house