Six AI leaders sign a voluntary White House accord on super intelligence safety

Six of the most powerful people in artificial intelligence have signed a one page safety pledge at the White House, and none of them can be punished for breaking it. President Donald Trump posted the voluntary accord on September 30, SiliconANGLE reported, and it commits the companies to internal controls, outside audits and board oversight of their most capable AI systems.

The document is titled the "White House Accord on Super Intelligence." It arrived alongside an executive order telling federal agencies to call AI "super intelligence" in official documents. A few days later the president announced a new federal task force, the Super Intelligence Force, to coordinate government work on the technology.

Who signed the White House accord

According to SiliconANGLE, the signatories are:

• Jensen Huang, chief executive of Nvidia.
• Sundar Pichai, chief executive of Google.
• Mark Zuckerberg, chief executive of Meta.
• Elon Musk, representing xAI.
• Dario Amodei, chief executive of Anthropic.
• Greg Brockman, president of OpenAI.

The list covers most of the companies building the largest AI models in the United States, plus the dominant maker of the chips those models run on. OpenAI is represented by its president rather than its chief executive. Nvidia's presence is notable for a different reason: it does not mainly sell chatbots to the public, but nearly every frontier lab depends on its hardware.

What they agreed to

SiliconANGLE described four commitments:

• Internal controls that track what models can do in sensitive areas such as biology, and systems to block cyberattacks.
• Dedicated safety teams, with outside auditors checking that the controls work.
• Independent board committees that oversee those teams and review audit reports.
• Regular meetings between the companies to set shared safety standards.

CBS News reported that Trump called the standards "morally binding" and said the companies "understand that they have to self-police."

What those commitments mean in practice

Each item maps onto work that large labs already describe doing in some form.

Capability tracking is the practice of testing a model before release to see what it can do in high risk areas. For biology, that usually means checking whether a model gives meaningful help to someone trying to cause harm, beyond what is already easy to find. For cyber, it means testing whether a model can find or exploit software flaws. The goal is to know where a model sits before deciding how to release it.

Safety teams and outside auditors separate the people building a product from the people checking it. An internal team runs the tests. An outside auditor confirms the tests were done properly and that the controls hold up. The value of the auditor depends entirely on independence and on what the auditor is allowed to see.

Board committees move responsibility up the chain. If a board committee receives audit reports, directors cannot easily claim they did not know about a problem.

Shared standards are the most open ended item. Regular meetings could produce common testing methods, which would make results easier to compare across companies. They could also produce very little. The accord, as described, does not set a timetable for any output.

What the accord does not do

The accord is voluntary. No penalties are written into it, and no government body has been given the power to enforce it. The signatories themselves wrote, according to SiliconANGLE: "We believe every company is responsible for developing its own technology safely."

That sentence frames the whole arrangement. Responsibility stays with each company, and the White House acts as host and witness rather than referee. There is no stated requirement to publish audit results, no named auditing body and no process for a member of the public to raise a concern.

Voluntary pledges are not new. Major AI companies made a similar set of White House commitments in 2023, covering testing, security and transparency, and those also relied on companies keeping their word. The question then, as now, was how anyone outside would know whether the promises were kept.

Who it affects

For the companies, the accord is low cost and high visibility. Signing costs little, and it gives each one a public commitment to point to when regulators or customers ask about safety.

For businesses that build on these models, the main effect may be indirect. If the controls are real, they could mean more testing before release and, at times, delays. A recent example came the same week, when OpenAI held back its GPT-6.1 Astra model after internal safety tests. That decision was made under OpenAI's own process, not under the accord, but it shows what an internal check can do when it is taken seriously.

For ordinary users, nothing changes this week. The products you use will look the same tomorrow. Any benefit would show up over time, as fewer harmful capabilities slip through and fewer rushed releases reach the public.

What to watch next

Voluntary pledges can shape how companies test and release AI tools, but they depend on the companies following through. The useful test over the next year is whether any of the promised outside audits are published, and whether a company ever delays a product because an audit flagged a problem.

A few other signals are worth tracking:

• Whether any company names its outside auditor.
• Whether the shared standards meetings produce a public document.
• Whether the new federal task force asks to see audit reports as part of its 120 day review.
• Whether other AI developers, beyond these six, sign on.

We have made the case that voluntary AI safety pledges need public proof before they earn trust. This accord will be a clean test of that argument. For more on how the policy story is unfolding, see our AI coverage.

Frequently asked questions

Is the White House accord legally binding?

No. SiliconANGLE reported that it is voluntary, carries no penalties and gives no government body power to enforce it. Trump described the standards as "morally binding," according to CBS News, which is a political statement rather than a legal one.

Which companies signed?

Nvidia, Google, Meta, xAI, Anthropic and OpenAI, represented by Jensen Huang, Sundar Pichai, Mark Zuckerberg, Elon Musk, Dario Amodei and Greg Brockman, according to SiliconANGLE.

Will audit results be made public?

The accord, as reported, commits companies to outside auditors and board review of audit reports, but it does not say the results will be published. That is the biggest open question.

Does this change how I use AI apps?

Not right away. Any effect would come gradually, through more testing before new models are released and, possibly, occasional delays when tests turn up problems.