Donald Trump has brokered a deal with two dozen tech firms to police themselves on AI safety, a plan that relies on voluntary agreements rather than new laws. The companies signed a letter committing to independent safety audits, regular meetings to share best practices, and a pledge to work together on AI standards. The deal, announced Tuesday, has no legal force — but Trump called it “morally binding” anyway.
The agreement covers a wide range of risks, from cybersecurity to biosecurity to chemical threats. It also addresses unintended model actions, the kind of behavior that can emerge when an AI system acts outside its intended boundaries. The companies agreed to test whether their internal controls, monitoring, and detection systems actually work.
Who Signed the Agreement
The signatories include some of the most powerful figures in tech:
- Dario Amodei, of Anthropic
- Sam Altman, of OpenAI
- Elon Musk, through SpaceXAI
- Jensen Huang, of Nvidia
- Mark Zuckerberg, of Meta
- Sundar Pichai, of Alphabet and Google
These six leaders represent the companies driving the AI industry forward. Their signatures carry weight, but the question remains: weight of what?
What the Deal Actually Says
The letter commits firms to a few key steps. They will conduct independent safety audits, which test whether internal controls actually work. They will hold regular meetings to share best practices and set common AI safety standards and benchmarks. And they will consider whether to codify the recommendations later.
The agreement states that it “may make sense to codify” the recommendations. That is a hedge, not a promise. The deal is voluntary, and the companies are free to drop out at any time.
Some firms have already met with the Center for AI Standards and Innovation, which runs voluntary government pre-deployment safety tests. Many firms have voluntarily committed to highly secretive government safety testing. The deal simply formalizes what some were already doing.
The Limits of Self-Policing
The problem is that the deal has no teeth. There is no enforcement mechanism. There are no fines. There is no legal requirement for compliance. The companies agreed to do these things, but there is nothing stopping them from failing to do them.
Trump called the agreement “almost like a constitution, in a way.” He told reporters at the White House: “The biggest people in the world signed that, and I signed it as president, and it really is a form of protection.”
That is a strange way to describe a voluntary agreement. A constitution is the supreme law of a country. It carries legal force. It can be enforced. The deal Trump described is none of those things.
Some AI firms are uncertain what qualifies as a third-party tester. The selection process appears to be politicized, according to Bloomberg. It is unclear who Trump is tapping to audit frontier models.
The Companies Already Committed
Not all of the signatories are new to this game. Some firms had already committed to external audits before this deal was announced. Anthropic, Google, and OpenAI were among those who had made prior public commitments.
That raises a question: what does this deal actually add? The companies who were already auditing themselves are now signing a letter promising to continue doing what they were already doing. For them, the deal is largely symbolic.
For the companies that have not yet committed to external audits, the deal creates a norm. If enough firms agree to the same standards, the industry moves in a particular direction. Whether that movement is faster or more effective than regulation is an open question.
A Comparison of Commitments
| Company | External Audits | New Commitment |
|---|---|---|
| Anthropic | Previously committed | Signed letter |
| OpenAI | Previously committed | Signed letter |
| Previously committed | Signed letter | |
| Nvidia | Unclear | Signed letter |
| Meta | Unclear | Signed letter |
| SpaceXAI | Unclear | Signed letter |
The table shows that three of the six signatories had already committed to external audits before the deal was announced. The remaining companies did not disclose their prior positions publicly.
What Comes Next
The deal is a statement of intent. It is a way for the companies to signal to each other and to the public that they are taking AI safety seriously. It is also a way for Trump to show that he is engaged with the tech sector.
But the hard work is ahead. Independent safety audits require independent auditors. Those auditors need to be credible, competent, and free from conflicts of interest. The deal does not specify how auditors will be selected or certified.
The regular meetings are also a commitment. Sharing best practices is valuable, but it is not the same as setting mandatory standards. The benchmarks the companies develop could become de facto standards — but only if everyone agrees to use them.
There is a risk that the deal becomes a talking shop. Companies meet regularly, exchange ideas, and pat each other on the back. Nothing changes in practice.
A Question of Trust
The deal rests on trust. The companies trust each other to follow through. The public trusts that the companies will actually conduct meaningful audits. Trump trusts that the companies will protect themselves and the country.
That is a lot of trust to place in a voluntary arrangement. It is also a lot of trust to place in an administration that has shown little appetite for regulating tech firms.
The deal is a start. It brings together the companies that matter most in AI. It puts safety front and center in their discussions. But it is not a solution.
A solution would involve actual legal requirements, backed by enforcement. A solution would involve public oversight, not just private agreements. A solution would involve transparency, so the public knows what is being tested and how.
The deal is a step. It is not the destination.
Where the paper stands
The paper backs narrow rules against direct harm, like forcing companies to disclose safety failures they hid, and is against broad rules that hand the market to the incumbents, including deals that let big companies police themselves without any legal teeth. The Trump deal announced Tuesday fits neither category. It is a voluntary arrangement with no enforcement, no fines, and no legal requirement for compliance — a constitution without the force of law. The paper opposes this approach because it locks in today’s leaders while leaving the public unprotected.
The signatories — Dario Amodei of Anthropic, Sam Altman of OpenAI, Elon Musk through SpaceXAI, Jensen Huang of Nvidia, Mark Zuckerberg of Meta, and Sundar Pichai of Alphabet and Google — carry weight, but the weight is moral, not legal. The paper supports companies disclosing safety failures they hid, but it opposes letting companies police themselves without any legal teeth. The deal offers neither.
The companies that were already auditing themselves are now signing a letter promising to continue doing what they were already doing. For them, the deal is largely symbolic. The remaining companies did not disclose their prior positions publicly, and the selection process for auditors appears to be politicized. The deal may make sense to codify the recommendations, but it is a hedge, not a promise. The hard work — credible, competent auditors free from conflicts of interest — remains undone.
Source material: “Trump plan to combat AI risks hinges on Big Tech pals policing themselves,” Ars Technica.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.