Skip to content

Leadership6 publishersWidely confirmed3 min readPublished Updated

Trump's 'morally binding' AI pact lets the labs choose their own auditors

Donald Trump and top US AI chiefs signed a voluntary pact with four layers of safety controls, none involving a government regulator. Any accountability a buyer can enforce still has to come from its own contracts and testing.

The Board Room · Leadership desk

How we use AISend a correction

Photograph accompanying Trump's 'morally binding' AI pact lets the labs choose their own auditors
Photo: theguardian.com

What happened

  • The pact does not commit any company to publish the findings of independent evaluations of its safety controls.
  • According to the Guardian, the first layer, internal monitoring during training, already exists at major labs and has not prevented testing failures.
  • A separate executive order renames AI as Super Intelligence for federal use and gives the science adviser 60 days to propose a legal definition.

Compiled by The Board RoomSomething wrong?How this is made

Why it matters

  • exposure A deployer relying on a signatory's safety claims has no published audit finding to check them against, so its own due diligence is the only independent check it can see.
  • decision Renewals this quarter are where a buyer can require a named auditor and access to board-reviewed reports, because the pact leaves both with the vendor.
  • precedent With the president's signature on it, a self-selected audit becomes the answer a vendor can give when a customer asks who checks its models.
  • constraint Vendor contracts written against today's federal AI definition stay valid, so any change to legal terms waits on the adviser's proposal.

Each of the pact's four layers can be sorted by who runs it. The first is internal safety monitoring during training, and the second is an internal team that checks the first is working [3]. The third lets an external auditor assess the safety controls, and the fourth sends reports to an independent board [3]. According to the Guardian, the agreement appears to let companies pick their own evaluators and appoint their own oversight boards [4]. That puts two layers inside the company and the other two under its selection [18]. The Guardian also reported that the first layer already exists at major labs and has failed to prevent testing failures so far [9].

"It's almost like a constitution in a way, and the biggest people in the world signed that, and I signed it as president, and it really is a form of protection," Trump said of the pact [5]. His description leaves out what a buyer needs to know. None of the controls include a commitment to publicly detail the findings of independent evaluations, and companies appear to decide for themselves whether to publish results [13][4]. House Speaker Mike Johnson called the document a "statement of principles" that will voluntarily hold the companies to "robust internal controls" [8].

The strongest case for the pact came from the executives at the lunch, who included Musk, Zuckerberg, Huang and Amodei [11]. "This is a start and an accord that the whole industry could come to," Zuckerberg said [14]. Amodei said: "We all need to work together to make sure that we can win, and we can win safely if we do this right" [15]. A customer's contractual position is what it was before Tuesday's lunch. The Guardian found no apparent enforcement mechanism or legal implications in the commitment [1]. Asked whether it was binding, Trump said: "I think it's morally binding" [7].

Trump has broadly rejected calls for more regulation of the technology, citing competition with China and threats to economic growth [10]. The nearest federal deadline in Tuesday's record belongs to the executive order renaming AI "Super Intelligence" [12]. It gives the president's science and technology adviser 60 days to propose legislation defining the term. For now the label covers the same technologies as the existing federal definition [16]. Agencies will not have to rewrite regulations or contracts already issued [17].

In my view, the standard a deployer can rely on this quarter is the one in its own contract. The pact leaves the choice of evaluator, the board appointments and the publication decision with the labs [4]. A buyer can ask a vendor to name its auditor and share what its board reviewed, and make both a condition of renewal. A renewal signed without them this quarter is the contract the buyer will hold if its vendor has a testing incident next quarter. The industry has already had such incidents, in which major AI companies inadvertently hacked into outside organisations during safety tests [6].

What to watch

  • The science and technology adviser's proposed legislation defining Super Intelligence, due within 60 days of the order.
  • Whether any signatory names its external auditor or publishes an independent evaluation finding.
  • A further testing incident at a signatory, and whether its independent board's review is made public.

Clarity's read

What the record supports and how the coverage leans. The claims behind it follow.

Reality

Evidence78
Adoption15
Hype gap+55
Incentives75
Confidence70

Perspective Coverage

6 publishers
Builder
Builder 23%
Operator
Operator 45%
Investor
Investor 32%
Why these scores

Claim ledger

Ranked by verification strength, evidence, and original report placement.

  1. [1]

    The joint commitment Trump posted to Truth Social appears to carry no enforcement mechanisms or legal implications and is instead a voluntary commitment.

    ReportedSupportedSource: The Guardian's reading of the document5 sources— create a free account to open themView cited source
  2. [2]

    Donald Trump announced on Tuesday that the heads of the largest US tech and AI companies had signed a 'morally binding' agreement called the 'Joint Commitment On Frontier Responsibilities', following a White House luncheon with top tech executives.

  3. [3]

    The document outlines four 'layers of controls and audits': internal safety monitoring during training; an internal team empowered to make sure the first layer is operating correctly; an external auditor able to independently assess safety controls; and an independent board designated to review reports about those controls.

Sources

6 independent publishers whose own reporting we read for this story.

  1. businessinsider.com

    5 articles · September 30, 2026

    Trump revealed the seating chart for his big AI meeting. See who's in — and who's missing.
  2. cbsnews.com

    1 article · September 29, 2026

    Trump and major AI executives sign "morally binding" voluntary controls: "It's almost like a constitution" - CBS News
  3. hcamag.com

    1 article · September 30, 2026

    Trump says AI firms will self-police, so employers should too
  4. implicator.ai

    3 articles · September 30, 2026

    Trump Orders Agencies to Call AI 'Super Intelligence'
  5. koreatimes.co.kr

    1 article · September 29, 2026

    Trump touts AI boss pledge to self-regulate - The Korea Times
  6. theguardian.com

    2 articles · September 30, 2026

    Trump announces vague ‘morally binding’ AI deal among tech CEOs for ‘tremendous self-policing’ | Donald Trump | The Guardian

Share your take

Let Clarity write the post for you.

Signed-in readers get a short post drafted on this story in the register they choose — narrative, analytical, or a direct position — editable to the last word before it goes anywhere. The share buttons at the top of this story work without an account.

Topics and entities

Follow any of these and your For You feed starts watching them — no settings page required.

Loading related stories