President Donald Trump and executives from six leading artificial intelligence companies have signed a voluntary framework for managing risks associated with frontier models, placing internal controls, external evaluation and board oversight at the center of the arrangement. The White House Accord on Super Intelligence creates a four-layer corporate oversight model without establishing legally enforceable obligations or penalties. Trump described the September 29 agreement as “morally binding” after meeting technology executives in Washington.
The accord was signed by Anthropic CEO Dario Amodei, Google CEO Sundar Pichai, Meta CEO Mark Zuckerberg, OpenAI President Greg Brockman, Nvidia CEO Jensen Huang and Elon Musk, founder of xAI, alongside Trump. The framework asks the companies developing frontier systems to assume primary responsibility for monitoring their own models while adding multiple review layers. AP reported that the document explicitly leaves open the possibility that some measures could later be codified into laws or regulations.
Accord Builds Four Layers of AI Oversight
The first layer calls for robust internal controls designed to ensure frontier models behave as intended, including monitoring for cybersecurity, biosecurity and other potentially dangerous capabilities. Companies are then expected to maintain dedicated teams that verify those safeguards and remediate identified problems. The structure treats continuous internal monitoring as the first line of defence rather than prescribing a common technical safety standard across the industry.
A third layer requires companies to engage an independent external auditor or evaluator, while a fourth calls for an independent committee of each company’s board to review findings and oversee remediation. External review adds separation from the teams building the models, but the accord does not specify common auditor-selection rules, accreditation requirements or mandatory disclosure of findings. Those unresolved details will determine how comparable and independently verifiable company implementations become.
Similar control questions are already emerging where AI systems interact directly with digital assets. Bitcoin infrastructure has experimented with bounded authority for AI agents operating wallets, while payment systems are developing tokenized credentials and spending limits for autonomous agents. Those implementations illustrate the broader governance problem: autonomy becomes more consequential when software can access systems, credentials or financial assets without continuous human intervention.
Voluntary Model Leaves Enforcement Questions Open
Trump has framed the approach around maintaining U.S. technological leadership without imposing restrictions that he argues could slow development. Reuters reported that he said the companies must “self-police,” while AP cited experts who differed over whether voluntary corporate controls provide sufficient public accountability. The central policy distinction is that the accord establishes commitments and review processes, not a new federal regulatory regime.
The agreement also comes as AI developers themselves increasingly acknowledge security risks from more autonomous systems. Verification is becoming relevant beyond model laboratories as well, with Ethereum researchers exploring AI-assisted software verification alongside longer-term security work. Meanwhile, investment in infrastructure for autonomous AI agents is expanding the number of economic systems where authorization and monitoring controls could matter.
Separately, Trump signed an executive order on September 29 directing executive agencies, to the maximum extent permitted by law, to use “Super Intelligence” and “SI” instead of “Artificial Intelligence” and “AI” in specified federal communications. That terminology directive is legally distinct from the companies’ voluntary safety accord and does not provide an enforcement mechanism for it. The order also requires proposed legislative language on a federal definition of “Super Intelligence” within 60 days.
The next concrete milestone is therefore implementation rather than signature. Publication of audit methodologies, evaluator-selection procedures, board reporting structures and remediation standards would show how the voluntary commitments operate in practice, while any subsequent legislation codifying the framework would represent a materially different level of government oversight.
