Trump has secured a voluntary AI safety deal with Google, Anthropic, Meta, OpenAI, xAI and Nvidia.
The “morally binding” agreement does not have the force of federal law or impose new penalties.
The deal calls for stronger AI safeguards, independent audits and board-level oversight.
US President Donald Trump has secured a voluntary commitment from leading artificial intelligence companies to police their own systems, offering a White House-backed alternative to immediate government regulation as concerns grow over increasingly autonomous AI.
Executives from Google, Anthropic, Meta, OpenAI, xAI and Nvidia signed the accord after meeting Trump at the White House on Tuesday. Trump described it as “morally binding” and said the companies understood they needed to “self-police”. But the agreement does not have the force of federal law and does not set out penalties for companies that fail to meet its commitments.
Instead, the one-page accord asks companies to strengthen internal safeguards, use independent external auditors and give their boards a role in reviewing those assessments. It also leaves open the possibility that some of the measures could eventually be written into laws or regulations.
What have the AI companies agreed to?
The agreement sets out four layers of control. Companies are asked to work with independent external auditors or evaluators, establish an independent board committee to review reports from auditors and AI-control teams, maintain internal controls to monitor the capabilities and alignment of AI models during training and deployment, and have an internal team check that those controls, monitoring and detection systems are working as intended.
The accord also calls for internal controls to monitor AI models during training and deployment for risks including cybersecurity, biosecurity and chemical threats, and to ensure the models do not “hack or access technical systems in unintended ways”.
That provision matters as AI developers move towards systems capable of performing tasks with limited human intervention. AI agents can interact with software, use digital tools, write and execute code and take actions on behalf of users, creating risks beyond inaccurate answers.
The aim is to put safeguards around those capabilities while subjecting the companies developing them to additional scrutiny.
The document also leaves the door open to government action. It says that, “over time”, it may make sense to codify some of the measures into laws or regulations.
What does ‘morally binding’ actually mean?
In practical terms, “morally binding” describes a public commitment backed by political and reputational pressure rather than a legal obligation.
If Congress passes a law requiring an AI company to meet a particular safety standard, the government can enforce that requirement through the legal system and impose penalties where the law allows. The White House accord does not create that kind of enforcement mechanism.
The companies have publicly agreed to follow the principles, but the accord itself creates no new offence or financial penalty for failing to do so.
AP reported that Trump described the agreement as voluntary while saying he was seeing “tremendous self-policing” among the companies. He has argued that the US should not “stifle” AI development with regulation while maintaining its lead in the technology.
Trump also described the agreement as “almost like a constitution”, portraying it as a common framework for the companies involved. But the comparison has limits. A constitution or statute establishes a formal legal framework; this accord does not.
The companies may face political, investor and reputational pressure to comply, but those consequences are different from legal enforcement. The accord itself does not give the government new powers to fine a company or block a model for failing to meet one of its commitments.
Why has AI self-regulation become an issue now?
The agreement comes as concerns grow over AI systems that can operate with greater autonomy.
Al Jazeera reported that the accord came amid growing scrutiny of AI risks following incidents involving AI agents from OpenAI and Anthropic. The companies have reported cases in which their AI agents went rogue and hacked into other companies' systems, raising questions about whether safeguards can keep pace with increasingly autonomous AI.
When an AI system can act on its own, a failure can involve a real computer system, network or piece of data.
A Reuters/Ipsos poll conducted from September 17 to 20 found that 73 per cent of respondents were concerned that AI companies had not done enough to prevent serious harm from the technology, while 55 per cent said it would be good to slow AI development.
At the same time, Trump has continued to support rapid expansion of the US AI industry. Reuters reported that the White House meeting also covered the expansion of data centres needed to develop and operate increasingly powerful AI systems. Those facilities have faced opposition in some communities because of their electricity consumption and other local effects.
Self-regulation allows Trump to address some of those concerns without immediately imposing a new federal AI regime.
What happens if self-policing is not enough?
Trump has also floated a separate 10-person board to oversee AI safety more broadly.
That proposed body should not be confused with the committees that individual companies have agreed to establish. The company committees would operate within each business and review reports from internal and external auditors. The 10-person board Trump is considering would be separate from those company-level committees, although its membership and powers have not yet been defined.
Reuters reported that Trump is considering such a board and plans to appoint a new White House official responsible for AI policy. That suggests the administration is exploring government oversight while continuing to rely on voluntary commitments from the industry.
The key question is what happens after the companies sign it.
An independent audit can identify weaknesses. A company board can review those findings. But the voluntary agreement does not establish a new government penalty if a company fails to address a problem identified by an auditor.
The accord also does not settle the wider debate over whether AI companies should ultimately face mandatory safety requirements. Instead, it leaves that possibility open by acknowledging that some of its provisions could eventually be incorporated into legislation or regulation.
For Trump, the arrangement allows him to emphasise AI safety without adopting the stricter regulatory approach he has opposed. For the companies, it offers a way to demonstrate that they accept external scrutiny while continuing to develop their systems.
The unresolved question is whether voluntary safeguards can keep pace with increasingly capable AI systems.
If companies implement the controls, accept meaningful independent scrutiny and act on problems identified by auditors, the accord could become a basis for broader AI safety standards. If they fail to do so, pressure for enforceable regulation could grow.
For now, “morally binding” means a public commitment backed by political and reputational pressure, rather than an obligation backed by law. The companies have agreed to police themselves; whether that is enough will depend on how those safeguards work in practice and whether Washington eventually decides that voluntary measures are insufficient.
