Frontier AI · Trust & policy
Trump’s Super Intelligence Accord: Four Safety Promises From Six AI Companies
The signed document calls for internal controls, outside assessment and board oversight. The next test is how those promises are implemented.
President Donald Trump published the signed White House Accord on Super Intelligence on September 29. Subtitled Joint Commitment on Frontier Responsibilities, it sets out company-led controls and review processes for frontier AI. The original post contains the commitment and signature pages.
The signature page names Trump alongside Sundar Pichai of Google, Dario Amodei of Anthropic, Mark Zuckerberg of Meta, Greg Brockman of OpenAI, Elon Musk of xAI and Jensen Huang of Nvidia.
The four commitments
The document asks companies training and deploying frontier models to put four layers of oversight in place:
| Layer | Commitment |
|---|---|
| Internal controls | Monitor model capabilities and alignment during training and deployment, including cyber, biological and chemical risks and unintended access to technical systems. |
| Internal review | Give a team responsibility for checking that controls work and problems are corrected. |
| Outside assessment | Work with an independent external auditor or evaluator to assess those controls. |
| Board oversight | Establish an independent board committee to receive reports and oversee remediation. |
Participants also commit to meeting regularly on standards and best practices. The text leaves open the possibility of putting these measures into future laws or regulations. Source: the signed accord.
The implementation questions
The published text does not supply a common implementation deadline, a numerical threshold for which models count as frontier systems, or a requirement to publish audit findings. It does not name the external evaluators. Those are details to watch in follow-up announcements.
Our analysis: the strength of this arrangement will depend on what reviewers can inspect and what happens after they find a problem. An evaluator needs enough access to test a company’s claims. A board committee needs evidence that problems were resolved. Readers should look for those operational details when companies describe their participation.
A useful follow-up would identify the review’s scope, the model versions it covered and who was responsible for closing the findings. A public summary could help customers understand the results without disclosing sensitive technical information. These are questions raised by the announcement, rather than additional promises contained in the accord.
What Canadian AI buyers can ask
For an Alberta business choosing a model or an AI agent, the announcement offers a reason to ask more specific vendor questions. Which evaluation covers the version we use? What can the agent access in our environment? Who receives an incident report? Which actions still require our approval?
Our practical reading is that a supplier’s commitment should sit alongside a buyer’s own workflow review. A test of a frontier model does not tell a company whether its particular connection to email, customer records or payment tools is configured appropriately.
For a separate, established reference, the NIST AI Risk Management Framework is voluntary guidance for incorporating trustworthiness into the design, development, use and evaluation of AI systems. It is background for those operational questions, not an endorsement or implementation plan for this accord.
What comes next
The next useful reporting will be company-specific: who is reviewing which systems, when the review happens and what evidence customers can see. The accord records a shared commitment. Its practical effect will become clearer as those details are published.
Sources and editorial note
- Donald J. Trump: original September 29 post, including both pages of the signed accord
- NIST: AI Risk Management Framework, background on voluntary organizational risk management
Both pages of the original post were checked on September 29, 2026. Statements about the document are reported from that source. Sections marked as our analysis offer editorial interpretation. The cover is an AI-generated illustration.