Trump's AI safety pact is voluntary, not binding
- Leading AI companies agreed at a White House meeting on 29 September 2026 to voluntary safety standards including independent audits.
- The commitments are not binding in law, and Trump said he was considering a ten-person board to oversee AI safety.
- A Reuters/Ipsos poll found that 73 per cent of respondents believe AI companies have done too little to prevent societal harm.
- Community opposition to data centre expansion over electricity costs has created political risk for Republicans ahead of the 3 November midterm elections.
On 29 September 2026 the White House announced that leading artificial intelligence companies had agreed to voluntary safety standards at a meeting with Donald Trump. Attendees included Greg Brockman of OpenAI, Dario Amodei of Anthropic, Mark Zuckerberg of Meta, Sundar Pichai of Google and Jensen Huang of Nvidia, alongside House Speaker Mike Johnson. A copy of the agreement, posted on Trump’s social media platform, commits the firms to work with independent auditors to check that their systems behave as intended, and to seek to prevent their tools from accessing technical systems in unintended ways.
Reuters’s report on the White House meeting and the safety accord described the commitment as voluntary, a word doing most of the work here. Nothing in the announcement binds the companies by law. The standards will be met or missed depending on what the firms choose to put before auditors, and on what those auditors find. Trump said he was considering a ten-person board to oversee AI safety and would soon appoint a new White House AI policy official. On the same day, federal agencies were directed to use the phrase super intelligence instead of artificial intelligence, a change in official vocabulary with no evident effect on how the systems behave.
That timing lines up with public scepticism already on record. A Reuters/Ipsos poll found that 73 per cent of respondents believe AI companies have done too little to prevent societal harm. Trump also repeated his support for rapid data centre expansion, arguing the companies would contribute substantially to host communities. Community opposition to data centres over electricity costs and other impacts has created political risk for Republicans ahead of the 3 November midterm elections.
Schools and universities that permit AI tools with conditions on declaration and appropriate use depend on evidence that the systems behave as described. Independent auditing is the kind of mechanism that could supply such evidence. A voluntary pact, though, leaves the scope, depth and disclosure of any audit to the companies themselves. Institutions are left with little more to verify than the companies’ own account of their practices. Officially relabelling the field as super intelligence adds a rhetorical layer that complicates guidance for students and staff rather than clarifying it.
Three things will be testable in the period after the meeting: whether the ten-person board is created, whether audit findings reach the public rather than staying with the companies, and whether promised benefits to host communities materialise. The 73 per cent figure suggests that many members of the public expect something firmer than an accord firms may honour on their own terms.