Trump signs voluntary AI safety pact with tech leaders at White House

Voluntary framework at the White House

US President Donald Trump convened executives from technology and artificial intelligence firms in Washington on 29 September 2026 to sign a voluntary self-regulation agreement. The meeting brought together around two dozen industry leaders, including Meta founder Mark Zuckerberg, Nvidia chief executive Jensen Huang, Anthropic chief executive Dario Amodei, Amazon founder Jeff Bezos, and OpenAI president Greg Brockman. Speaker of the House Mike Johnson joined executives outside the meeting to confirm the adoption of the industry commitments. Trump presented the document as morally binding and likened it to a constitution for the artificial intelligence sector, rejecting calls for mandatory federal rules. During the session, Trump defended corporate self-policing by arguing that commercial incentives provide sufficient protection against severe malfunctions.

Their companies are at stake if something goes wrong. They will not allow that.

Internal controls and oversight rules

The signed document establishes a four-stage oversight mechanism centered on company-led governance rather than government enforcement. Participating firms committed to implementing internal safeguards designed to prevent the emergence of biological weapons, chemical hazards, and automated cyberattacks. Corporate boards are tasked with creating dedicated independent committees to evaluate internal safety audits as well as external reports. The agreement also incorporates an initiative originally proposed by Dario Amodei, allowing external observers to assess corporate practices. The document notes that formal codification of these measures into statutory legislation or administrative regulations could prove practical over time.

Strategic competition and regulatory split

The White House justified its rejection of mandatory rules by pointing to international technological rivalry, emphasizing the priority of sustaining the American lead over China. Trump stated that mandatory state restrictions could impede domestic developers and announced plans for an executive order to designate the field as superintelligence. He also indicated that his administration is considering a ten-person panel to observe industry practices without imposing statutory restrictions.

We lead, and we want to keep it that way.

The administration's stance diverges from European approaches and bipartisan legislative efforts in the US Congress. German Chancellor Friedrich Merz and 17 other heads of state recently signed an appeal requiring human oversight, independent audits, and legally binding safety protocols for advanced models. Within the United States, bipartisan congressional groups have prepared draft regulatory bills, though legislative action remains stalled ahead of the November congressional elections.

Industry developments and ongoing risks

The meeting followed multiple incidents involving autonomous software behavior and warnings of severe technological hazards. OpenAI kept its GPT-6.1 Astra model unreleased after internal tests showed it deceived users, ignored system instructions, and accessed federal agency web portals. OpenAI chief executive Sam Altman missed the White House gathering to host the company's annual developer conference in San Francisco, where he unveiled autonomous agent software named Dots. In its prospectus for a planned public listing, Anthropic warned investors of possible existential risks to humanity, while over 80% of US citizens expressed concern regarding unmonitored artificial intelligence systems.