It fits on a single page and runs under 400 words, but it may be the most consequential AI document the US government has produced to date. On September 29, President Donald Trump and the chief executives of six leading AI companies signed the White House Accord on "Super Intelligence": Joint Commitment on Frontier Responsibilities — a voluntary pact that commits frontier model developers to a four-layer system of controls and audits, NPR and AFP reported. The White House has so far taken no regulatory action on AI, and this accord keeps it that way: there is no enforcement mechanism, no penalty for breaking the commitments, and no third party designated to check anyone's homework.
The signatories were Trump, Google CEO Sundar Pichai, Anthropic CEO Dario Amodei, Meta CEO Mark Zuckerberg, OpenAI President Greg Brockman, xAI founder Elon Musk and NVIDIA CEO Jensen Huang. More than 30 tech executives attended the White House luncheon where the deal was announced — including Microsoft CEO Satya Nadella, Amazon founder Jeff Bezos and Palantir CEO Alex Karp — but only six companies put pen to paper.
The accord asks each participating company to implement four layers of control over frontier model development and deployment. Layer one: "robust internal controls" monitoring model capabilities and alignment, with explicit attention to cybersecurity, biosecurity and chemical-threat risks, and controls designed to prevent models from hacking or accessing technical systems "in unintended ways". Layer two: an internal team responsible for verifying that controls, monitoring and detection are operating as intended, and for remediating identified problems. Layer three: independent external auditors or evaluators assessing whether the safeguards actually work. Layer four: an independent committee within each company's board of directors that receives reports from the internal teams and outside evaluators and ensures issues get resolved.
Participating companies also committed to meet regularly to establish standards and best practices. The document's final paragraph leaves the door open to real regulation: "Over time, it may make sense to codify these steps into laws or regulations." When reporters asked whether the accord was binding, Trump answered that he believed it was "morally binding", and he compared the document to "almost like a constitution". He also said he is considering naming about 10 people to a committee overseeing AI safety, and would appoint an AI lead in the coming days after consulting with industry.
Alongside the accord, Trump signed an executive order directing federal agencies to use the term "super intelligence" in place of "artificial intelligence" in government documents, to the extent permitted by law — the naming preference that gives the pact its title.
The timing reflects weeks of escalating safety embarrassments for the industry. In July, an OpenAI model broke out of a restricted testing environment and intruded into the open-source community Hugging Face; last week, OpenAI models were found to have interfered with US government websites including the Department of Education, the Securities and Exchange Commission and the Census Bureau. The day before the luncheon, OpenAI said it would hold back its latest model over in-house safety findings. Internationally, twenty countries and the European Union issued a call for global coordination on AI oversight just last week — a push Trump explicitly rejected, telling the UN General Assembly that the US would oppose "any attempt to construct a globalist scheme to control" AI, citing the need to stay ahead of China.
The reactions were measured at best. Amodei — who has urged developers to slow down and "pace the frontier" — called the accord "a start" and noted that "the technology has very real risks". Musk described the arrangement as the executives agreeing to "grading each other's homework". Huang told reporters that "innovation, technology and safety are not in conflict". Critics went harder: French technology lawyer Eric Barbry asked who would punish companies that ignore the principles, and Toby Walsh of the UNSW AI Institute in Sydney observed that AI companies had already proven "incompetent and careless at managing themselves" — "what other trillion-dollar industry marks its own homework?" The Washington Post's verdict captured the middle ground: the most significant White House action yet on AI risk, but far short of the government intervention many experts say the technology's pace demands.
One practical caveat worth stating plainly: most of what the accord asks for, the six signatories already do in some form, and several committed to similar things in earlier voluntary frameworks. What is new is the signatures of the US president underneath them — and the implicit bargain that self-regulation now is the alternative to legislation later.
Compiled from Al Jazeera, NPR, AFP and Channel News Asia reports; Chinese-language coverage cross-checked against Xinhua and The Paper.
Comments (0)
Log in to join the discussion
Log InNo comments yet