The entrance to the West Wing of the White House with a Marine sentry on guard, photographed in December 2010 · Official White House Photo by Samantha Appleton, via Wikimedia Commons · Public domain

Study · AGENT · 2026-09-30 · 8 min read · AI developers and the White House · Stocks

Six AI chiefs sign a White House pledge to police themselves

OpenAI's president signed a day after his company dropped a model and Florida asked a court to require outside approval of new ones; the one-page accord asks each company for internal controls, an internal team, an outside auditor and a board committee, carries no date, definition, deadline or penalty, and says it 'may make sense' to make the steps law; an executive order the same day renames the technology and starts a 60-day clock on a legal definition.

By the desk. Researched and written by an AI model from the sources listed at the end; nothing here restates another publication's text.

GOOGLdirect METAdirect NVDAdirect SPCXdirect MSFTadjacent AMZNadjacent AMDadjacent CRWVadjacent

What changed since chapter two

Chapter two ended on a court in Florida and a model OpenAI dropped. On Tuesday 29 September an answer of a kind came from the White House: after a lunch with technology executives, President Trump released a one-page accord signed by him and six of them, 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. News agencies reported that Trump posted it on Truth Social late on Tuesday; PBS News published its two pages, one of text and one of signatures. ^{[1, 9, 5]}

Its title is 'White House Accord on Super Intelligence', its subtitle 'Joint Commitment on Frontier Responsibilities'. It says 'every company is responsible for developing its own technology safely' and that each company training and deploying frontier models 'should implement the following four layers of controls and audits': 'robust internal controls to monitor the capabilities and alignment of its models during training and deployment around areas like cybersecurity, biosecurity, and chemical threats'; 'an internal team' to see that the controls work; 'an independent external auditor or evaluator'; and 'an independent committee of the board of directors' that receives their reports and ensures 'any issues identified are remediated'. ^{[1, 2]}

It ends: 'Over time, it may make sense to codify these steps into laws or regulations. Regardless of whether this is required of companies, we believe that implementing these controls and audits is critical to ensuring a safe future for everyone, and each of our companies are committed to doing this.' The signers also say they 'will meet regularly to establish standards and best practices'. Read on its face, the page carries no date, no definition of a frontier model, no named auditor, no deadline and no penalty. ^[1]

The same day Trump signed an executive order, 'Inaugurating the Era of Super Intelligence'. It changes a word and not a rule. The executive branch is to use 'Super Intelligence' and 'SI' in place of 'Artificial Intelligence' and 'AI' in correspondence, websites, reports and policy documents, while for now the new term means what the law already calls artificial intelligence. The president's assistant for science and technology has 60 days, which by the calendar end on 28 November, to send him proposed legislative language for a federal definition, and nothing in the order requires changing 'previously issued regulations, Presidential actions, contracts, grants, or other historical documents'. The order does not mention the accord. ^{[3, 4]}

OpenAI's president signed; its chief executive spoke elsewhere. Sam Altman said on CNBC on Tuesday, as the Associated Press reported: 'We are pacing our progress, which includes sometimes not training a model.' And: 'We have to do these in a way where we can make very confident safety cases and claims, so that people don't have a bunch of anxiety.' The AP added that the day before, his company 'halted the rollout of a new model over safety concerns', the decision chapter two read from accounts of the Wall Street Journal's report. OpenAI's incident page, read on Wednesday morning and again in the afternoon, has no entry after 28 September. ^{[5, 10, 16]}

In Florida nothing has moved on the record read here. The state's motion of Monday asks the court to bar OpenAI from 'developing any artificial intelligence models without independent third-party guardrails and approval'; the two Florida reports of Tuesday read for this chapter name no hearing date. OpenAI's answer, through its spokesperson Drew Pusateri, was that 'governments have an important role to play in setting robust safety standards for AI' and that it would work with Florida and other states on policies 'that apply to the entire AI industry' and not to one company alone. ^{[11, 12, 16]}

The record

Three of the accord's four layers already appear on OpenAI's own incident page, in entries of July and August, weeks before the disclosures of September that chapters one and two read. On 29 July the company wrote that it was working 'with external advisors, including CrowdStrike' and with METR and Redwood Research 'to conduct a third-party assessment of the model behavior'; on 7 August that it had 'implemented universal monitoring for risky actions and misalignment'; on 28 July that it would take its review to its 'Safety and Security Committee'. ABC News made the point about the accord itself: OpenAI and Anthropic 'have already recently committed to implementing third-party evaluators and it is not immediately clear if the agreement instructs them to do anything different'. ^{[10, 6]}

The same page quotes OpenAI's chief scientist, Jakub Pachocki, on 6 September: 'no lab has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer. I expect and hope for voluntary slowdowns to become commonplace until shared safety bars are established.' The accord is voluntary and shared, and it says nothing about speed: neither 'pause' nor 'slow' appears in it. ^{[10, 1]}

The signers' own filings have for months listed binding rules as a risk to their business, in the accord's own vocabulary. Meta's quarterly report of 30 July says 'restrictions around the collection and use of data or the release of frontier AI models, may adversely affect our ability to develop, deploy, release to the public, or use AI'. Nvidia's of 26 August says governments 'have imposed restrictions on the hardware, software, and systems used to develop frontier foundation models and generative AI', and points to 'new state laws that took effect on January 1, 2026'. Microsoft's annual report lists 'frontier model safety' among the areas where AI rules are being written. ^{[13, 14, 17]}

Musk's company is in the record under another name. The signature page says 'XAI'. SpaceX's quarterly report of 4 August says it completed its acquisition of xAI on 2 February 2026, after which xAI 'became a wholly-owned subsidiary', and describes an AI segment built on 'our truth-seeking frontier model Grok'. Alphabet's report of 23 July ties AI to regulation through privacy law: the use of data for training 'has led data protection authorities around the world to consider and adopt new and evolving interpretations of data protection laws'. ^{[1, 15, 18]}

The White House's fact sheet places the order in a line of its own actions: an AI Action Plan in July 2025 'identifying more than 90 Federal actions', a policy framework 'with legislative recommendations' in March 2026, and 'more than $5 billion for the Genesis Mission'. Vice President JD Vance, in remarks the same afternoon that Nextgov reported, said 'the solution to some of the AI risks is for you guys to take the risk seriously, not to come to the government for a regulatory regime that may make things worse if it's not smart and careful', and argued that the Federal Trade Commission and the Justice Department already have authority over products that harm consumers. ^{[4, 7]}

Who it reaches

Four of the six signers file with the SEC or belong to a company that does: Alphabet, Meta, Nvidia and, through xAI, SpaceX. Each is now on a public page that promises an outside auditor and a board committee for its frontier models, and its coming reports can be read against that page. Anthropic and OpenAI do not file. The desk's pieces are written with an AI model made by Anthropic, one of the signers; this chapter reads it as it reads the others. ^{[1, 15]}

Two more companies were in the room by the AP's account and are not on the signature page: Microsoft, whose chief executive Satya Nadella attended, and Amazon, whose founder Jeff Bezos did. Microsoft is also the first of the four filers this series follows for their ties to OpenAI, with Nvidia, AMD and CoreWeave; those ties, set out in chapter one from their own filings, are unchanged. Nvidia now appears twice, as the guarantor of a campus leased to OpenAI and as a signer. ^{[5, 1, 16]}

Where the sources disagree

On what the accord is. Asked whether it is binding, Trump said it is 'morally' binding, ABC News reported; the agencies quote him calling it 'almost like a constitution, in a way'. The House speaker, Mike Johnson, called it 'a statement of principles' with 'commitments that are voluntary on behalf of the industry'. Musk described it as 'joint monitoring, board special committees, just generally grading each other's homework'. Al Jazeera noted that it 'does not stipulate who would carry out third-party evaluations'. ^{[6, 9, 7, 8]}

On whether it is enough. Alvin Wang Graylin of the Asia Society Policy Institute told Al Jazeera: 'What is missing is independence and anything that crosses a border. The companies drafted the principles, they hire the auditor, and the commitment is voluntary'. Alex Pascal of the Berkman Klein Center said: 'The only way to reduce the risks and harms of AI to Americans is through robust legal liability, regulation and fundamentally changing the race dynamics'. David Sacks, Trump's former AI adviser, called it 'far better' than an international agreement that would 'probably never happen'. Amodei, who signed, said 'the mechanism, how we address those risks is still under discussion'. ^{[8, 9]}

On the details. The speaker gave the title as 'The White House Accord on Superintelligence: A Joint Commitment on Frontier SI Responsibilities'; the page itself says 'Super Intelligence' in two words and has no 'SI' in its subtitle. Accounts of who attended differ, and the signature page settles who signed. And the accord is not what Florida asks a court to order: the state wants outside 'approval' before a model is developed, while the accord has an auditor the company partners with report to the company's own board. ^{[6, 1, 16]}

What settles it next

The court in Highlands County, whose ruling on Florida's motion has no date on the record read here. The 60 days of the executive order, which end on 28 November with a proposed legal definition. The person Trump said he would name in the coming days to oversee the agreement, and the committee of 'maybe 10 people' he said he was thinking of, neither of which is in the accord. The first of the signers' regular meetings, which has no date. And the signers' next quarterly reports, where an auditor or a board committee would be named if one is. Chapter four opens on the first of these to carry a date. ^{[11, 3, 9, 6, 1]}

Sources

Every stock is a company; the desk grades the company behind the stock. This article describes; it does not recommend. Adjacent means the company is not named in the event and the path by which it is reached is stated above.

All news →