US President Donald Trump and nearly two dozen tech executives have signed a voluntary accord that endorses using outside auditors to assess the safety of AI systems.
The agreement was reached at a White House lunch on Tuesday. AI companies also committed to meeting regularly to set industry best practices.
The document carries no legal weight, however, Trump told reporters it is "morally binding".
According to a copy Trump posted on Truth Social, signers include Anthropic CEO Dario Amodei, Nvidia CEO Jensen Huang and SpaceXAI's Elon Musk. OpenAI president Greg Brockman, Meta's Mark Zuckerberg and Alphabet CEO Sundar Pichai also signed.
"It's almost like a constitution, in a way," Trump said after the event.
What the Accord Says
The accord suggests AI companies "should" implement four layers of controls. It recommends internal controls for cybersecurity, biosecurity, chemical threats and unintended actions by AI models.
Companies should also build an internal team to monitor those controls. They should designate an independent committee of the board of directors to oversee safety teams and independent audits.
The document notes that "it may make sense to codify" its recommendations into law or regulation in the future. Several signers, including Anthropic and OpenAI, have previously committed to independent evaluators.
A Shift in Posture for the Administration
The embrace of third-party evaluators marks a change for the Trump administration. Officials have been sceptical of the groups best placed to run those reviews, over their ties to the effective altruism movement.
David Sacks, Trump's former AI czar, attended the lunch. He has dismissed outside assessments as "pseudoscience". Bloomberg previously reported that some of the firms poised to conduct reviews have already met the Center for AI Standards and Innovation, a unit inside the Commerce Department.
Trump has also previously called doomsday fears about AI "a hoax". AI has become politically charged ahead of the November elections, and a new Quinnipiac University poll found more than seven in 10 voters prefer candidates who back stricter AI guardrails.
The tone toward Amodei also shifted. Earlier this month, his pledge to slow cutting-edge AI development drew a rebuke from Trump on social media. On Tuesday, Trump said Amodei "has been fantastic" and called him forward to address reporters.
The two met for a private dinner on Sunday, hours after Trump rejected calls to slow AI development.
Amodei backed Trump's call to preserve the US lead in AI, but said the core safety questions remain unresolved. "The technology has very real risks," he said.
Limited Detail and Mixed Response
The administration held to its position that AI companies are responsible for policing themselves. Critics noted the accord's non-binding nature and limited detail. It does not specify who would carry out the third-party evaluations.
The images of Trump alongside billionaire executives could also give ammunition to opponents who accuse him of being too close to the tech industry.
Public concern has grown following an OpenAI agent's breach of an Australian government Medicare data portal. In July, OpenAI models also circumvented sandbox controls to access evaluation data hosted on Hugging Face. Warnings from former AI company employees that the technology poses an existential threat have added to the pressure.
OpenAI recently called on the US to lead global AI safety standards. Meanwhile, the US and China have been exploring ways to manage AI risks as the race for technological leadership continues.
Comments ( 0 )