# Trump and AI Company Chiefs Sign a Voluntary Accord to Police Their Own Models

The agreement commits Anthropic, OpenAI, Google, Meta, xAI and Nvidia to four layers of internal and external audits, and the president described it as only morally binding.

- Published: 2026-09-30T05:25:41.160Z
- Canonical: https://polylog.news/2026-09-30/trump-and-ai-company-chiefs-sign-a-voluntary-accord-to-polic
- Publisher: Polylog (Global desk)
- Section: tech
- Sources: [Al Jazeera](https://www.aljazeera.com/video/newsfeed/2026/9/30/trump-backs-ai-self-regulation-at-tech-summit-but-is-it-enough), [Euronews](https://www.euronews.com/2026/09/30/top-ai-and-tech-firms-sign-morally-binding-accord-to-self-police-development-after-meeting</source-url), [Financial Times](https://www.ft.com/content/9705f5bf-bc06-481d-b2a3-1299463074b8?syn-25a6b1a6=1)

President Donald Trump gathered the heads of the largest artificial-intelligence developers at the White House on Tuesday and announced a voluntary accord on safety controls. Al Jazeera reported that the president and leading technology companies [signed an agreement to regulate their own development work](https://www.aljazeera.com/video/newsfeed/2026/9/30/trump-backs-ai-self-regulation-at-tech-summit-but-is-it-enough). Euronews described the document as [morally binding rather than legally enforceable](https://www.euronews.com/2026/09/30/top-ai-and-tech-firms-sign-morally-binding-accord-to-self-police-development-after-meeting), a characterisation Trump himself used when asked.

The commitments are specific in structure and vague in enforcement. Executives from Anthropic, OpenAI, Google, Meta, xAI and Nvidia pledged four layers of controls and audits, including internal evaluations, review by an outside firm and oversight by each company's board, according to [Al Jazeera's account of the text](https://www.aljazeera.com/news/2026/9/29/trump-top-tech-firms-sign-accord-to-self-police-ai-development). The accord does not say who would conduct the third-party evaluations, and it states that codifying the measures into law [may make sense](https://www.cbsnews.com/news/trump-ai-constitution-tech-execs-openai-anthropic-voluntary-controls/) at some future point.

The pressure the accord answers is partly political and partly local. Euronews cited mounting demands from lawmakers and the public for rules after incidents involving autonomous systems, including OpenAI's disclosure that one of its agents accessed an Australian government website during testing. Separately, the Financial Times reported that the AI Infrastructure Coalition, whose members include Google, Meta and Microsoft, unveiled commitments to [cover energy costs and minimise water usage](https://www.ft.com/content/9705f5bf-bc06-481d-b2a3-1299463074b8?syn-25a6b1a6=1) as the industry tries to blunt opposition to data centres before the midterm elections.

Both moves have the same logic. Companies building capital-intensive infrastructure prefer to write their own compliance standards while they still can, rather than inherit rules drafted after an incident or an election.

## What this means

The mechanism is pre-emptive self-regulation aimed at delaying statutory liability, and the exposed parties are the developers themselves along with the utilities and municipalities that host their data centres. Voluntary audits impose real internal cost but avoid the discovery risk, injunction risk and approval delay that legislation brings. What is already expected is that this administration prefers industry commitments to rulemaking. What is new is the appearance of formal audit architecture and the promise to pay local energy costs, which concedes that the electricity and water burden of the buildout has become a voter issue. The precedent is the 2023 White House voluntary commitments, which were followed within a year by state-level statutes and litigation that went further than the pledges. The likelier outcome over the coming months is that state legislatures and courts, not Washington, set the binding constraints, and that compliance and grid-cost concessions raise the return the AI buildout must earn. A serious safety failure attributable to a signatory would compress that timeline and make federal legislation the base case instead.

## What to watch

- Whether any state attorney general or legislature moves on AI liability in the weeks after the accord, which would show the voluntary route did not stop binding rules.
- Whether the industry names actual third-party auditors, because an accord without identified evaluators leaves the audit layer unfilled.
- Whether utilities disclose who pays for new generation serving data centres, the concrete test of the energy-cost pledge.