Zuckerberg's draft became the 300-word AI safety accord six companies signed with Trump
Meta's Mark Zuckerberg shaped the 300-word AI safety accord that six companies signed with Trump, sources told Semafor. For investors, that makes the White House pact a commitment its signatories wrote for themselves, with federal rules left for later.
The Investor · Invest desk
What happened
- The pact began as a conversation between Zuckerberg and House Speaker Mike Johnson, who sat together at last week's state dinner for Xi Jinping, according to one source.
- Nvidia's Jensen Huang helped gather support for Zuckerberg's draft among the executives invited to Tuesday's meeting, including those from OpenAI, Anthropic and xAI.
- The accord commits the signatories to "robust internal controls" and to outside audits meant to ensure frontier AI models are safe.
- Some frontier models already go through a voluntary government clearance under a framework from a Trump order on AI that has never been made public.
Why it matters
- What it costs cost Near-term compliance spending under the pact depends on each signatory's own reading of what "robust internal controls" and an outside audit require.
- What it limits constraint Investors have no way to check a company's standing under government clearance while the framework behind it stays unpublished.
- What it sets up precedent If Congress writes the federal rules the accord says may be needed, the Speaker who first heard the idea from Zuckerberg would begin from principles drafted by the companies being regulated.
Next story · Security · Sep 30
Trump's AI accord leaves model safety checks to the model makersConfirmed5 publishers
Semafor's account has two layers of sourcing. The dinner origin rests on one source familiar with the matter, who also says Zuckerberg then talked to Nvidia's Jensen Huang about crafting principles that executives could use at their next meeting with Trump [15][16]. Two other sources confirmed Zuckerberg's central role in shaping the document, so three back that part [9][14]. Meta did not comment, and the White House did not return a request [13].
So Zuckerberg wrote the accord, or rather circulated the draft that Trump later released as the "White House Accord on Super Intelligence," a set of broad principles for self-regulation [7][4]. It is short. Just over 300 words across six signatories comes to about 50 words a company [3][1][12]. Semafor reports that the text does not get into significant detail about what the industry-run safety controls would look like in practice [3]. Zuckerberg called it a start. "The idea isn't that this is the only thing that we will ever do. It is that this is a start and an accord that the whole industry could come to," he told reporters outside the White House, standing next to Trump and Johnson [11].
I think the pact lowers the near-term regulatory risk an investor should assign to these six companies. The only binding piece in view is federal regulation, and the accord only hints that it may be needed in the future [5]. Trump has described enforcement elsewhere: an "AI Force" and an "AI Czar" to monitor bad actors, with the force working through the existing criminal justice system, according to a Benzinga report carried by Yahoo Finance [8].
The counter-thesis sits in the same 300 words. A document released under the president's name that commits its signers to outside audits hands a later rulemaker a standard the companies picked for themselves, and audits produce findings [2]. Public pressure could also break the view. Semafor notes it is not clear the gesture will quell broad public concerns about safety, and rules written in answer to that concern need not start from Zuckerberg's draft [6]. The view is wrong if the outside audits get named auditors and published results that a company can fail.
What to watch
- Whether the White House publishes the clearance framework under Trump's AI order, giving outsiders a standard to measure frontier models against.
- Whether Meta or the White House, neither of which commented to Semafor, confirms or disputes the account of how the accord was drafted.
- Whether AI developers beyond the six signatories put their names to the accord.
Information, not investment advice. This story is general information about companies and markets, drawn from cited reporting. It is not a recommendation to buy, sell or hold any security; check the cited sources before acting.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence72
- Adoption18
- Hype gap+45
- Incentives78
- Confidence66
Perspective Coverage
25 publishers- Builder
- Builder 20%
- Operator
- Operator 45%
- Investor
- Investor 35%
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
The accord was signed by Meta, OpenAI, Anthropic, xAI, Google and Nvidia.
- [2]
The accord lays out a commitment to 'robust internal controls' and outside audits designed to ensure the safety of frontier AI models.
- [3]
The agreement is just over 300 words and does not get into significant detail about what the industry-run safety controls might look like in practice.
- [4]
Trump released the document as a 'White House Accord on Super Intelligence,' outlining broad principles for self-regulation.
- [5]
The agreement hints that federal AI regulation may be needed in the future.
- [6]
It is not clear that the gesture will quell broad public concerns about safety.
- [7]
Zuckerberg circulated a draft ahead of Tuesday's meeting between Trump and AI executives; Huang played a role gathering support for it among the administration's guests, which included executives from OpenAI, Anthropic and xAI.
- [8]
Trump announced plans to establish an 'AI Force' and appoint an 'AI Czar' to monitor bad actors in the AI industry, with the force seeking out 'Bad' activity through the existing criminal justice system.
- [9]
Zuckerberg's central role in shaping the document was confirmed by two other sources familiar with the matter.
- [10]
Some frontier AI models abide by a voluntary clearance by the government under a framework established by a Trump order on AI, though that framework was never released to the public.
- [11]
The idea isn't that this is the only thing that we will ever do. It is that this is a start and an accord that the whole industry could come to.
- [12]
The accord's text works out to about 50 words per signatory.
- [13]
Meta did not comment for the story, and the White House did not return a request for comment.
- [14]
Three sources in all support Zuckerberg's central role: the one who described the dinner origin and two who confirmed it.
- [15]
The idea for Trump's Tuesday pact with the AI industry came from a conversation between Meta CEO Mark Zuckerberg and House Speaker Mike Johnson, seated next to each other at last week's state dinner for Chinese leader Xi Jinping, where Zuckerberg raised regulatory concerns about AI.
- [16]
Zuckerberg then talked to Nvidia CEO Jensen Huang about crafting a set of principles that tech executives could use at their next meeting with Trump.
Sources
24 independent publishers whose own reporting we read for this story.
- aljazeera.comTrump, top tech firms sign accord to ‘self-police’ AI development | News | Al Jazeera
1 article · September 29, 2026
- asia.nikkei.comTrump considers creating AI oversight committee after meeting with tech chiefs - Nikkei Asia
1 article · September 29, 2026
- cnbc.comZuckerberg, Amodei among tech executives set to meet Trump Tuesday
4 articles · September 30, 2026
- coindesk.comOpenAI, Google and Meta pledge independent AI safety audits under voluntary White House deal
1 article · September 30, 2026
- cointelegraph.comTrump accord calls for tech firms to ‘self police’ their own frontier AI
2 articles · September 29, 2026
- cryptobriefing.comNvidia, Anthropic, Palantir CEOs to meet Trump for AI risks lunch
6 articles · October 3, 2026
- cryptopolitan.comTrump and AI chiefs sign 'morally binding' safety accord at White House
1 article · September 30, 2026
- decrypt.coTrump Unveils 'Morally Binding' AI Accord Signed by OpenAI, Google and Nvidia
1 article · September 30, 2026
- dw.com'Super Intelligence': Why Trump wants to rename AI
1 article · September 30, 2026
- en.sedaily.comTrump Backs AI Self-Regulation, to Sign Order Renaming It Superintelligence
2 articles · September 30, 2026
- fastcompany.comWhat AI’s biggest CEOs really want from Washington
1 article · October 2, 2026
- finance.yahoo.comTrump to host Zuckerberg, Anthropic's Amodei, and other AI titans Tuesday
3 articles · September 30, 2026
- fortune.com‘We can’t stifle it’: Trump says AI leaders ‘have to self-police’ AI technology that ‘will be bigger than the industrial revolution’
2 articles · September 29, 2026
- foxbusiness.comPresident Trump orders federal agencies to replace AI with 'Super Intelligence' | Fox Business
1 article · September 29, 2026
- indianexpress.comTrump, tech leaders sign 4-tier pact for AI self-regulation | World News - The Indian Express
1 article · September 29, 2026
- livemint.comTrump vows to never stifle AI as he gathers with tech executives urging caution | Mint
3 articles · September 29, 2026
- nbcnews.comTrump to host summit with top AI leaders in Washington
1 article · September 29, 2026
- nytimes.comAt A.I. Event, Trump Asks Meta, OpenAI and Microsoft to Make Safety Decisions Themselves - The New York Times
1 article · September 29, 2026
- qz.comTrump hosts top AI executives at White House to discuss regulation
2 articles · September 29, 2026
- seekingalpha.comUnprofitable Superintelligence
1 article · October 1, 2026
- semafor.comAI leaders talk latest models, tech risks at Trump lunch
3 articles · October 1, 2026
- techtimes.comRogue AI Agents Hacked Healthcare Before Six Tech Giants Pledged to Self-Police AI Safety
1 article · September 30, 2026
- thehindu.comTrump, AI majors pledge AI safety; experts flag lack of enforcement - The Hindu
1 article · September 30, 2026
- usatoday.comTech leaders sign AI accord at White House, Trump says
1 article
- verdict.co.ukTrump and major tech CEOs sign voluntary accord on AI self-policing
1 article · September 30, 2026
Topics and entities
Topics
- Frontier AI safety
- Voluntary AI safety commitments
- AI Regulation