The Pulse
Six AI Firms Sign a Voluntary White House Safety Pact
OpenAI, Anthropic, Meta, Google, SpaceXAI and Nvidia signed a voluntary AI safety agreement with the Trump administration on September 29. The 308-word pact calls for safety protocols and internal monitoring teams, but leaves details of fed
AI.info Team · Written with AI ·
Six technology companies signed a voluntary AI safety pact with the Trump administration on September 29, but the 308-word agreement leaves key questions about oversight unanswered. OpenAI, Anthropic, Meta, Google, SpaceXAI and Nvidia committed to safety protocols and internal teams to monitor their own progress. The administration has not explained how it will check whether the companies follow through.
At a White House event after the signing, President Donald Trump called the arrangement “almost a constitution” for “Super Intelligence,” the administration’s preferred term for AI. He later told reporters, “I’m seeing tremendous self-policing.” The companies’ pledge comes as Washington favors industry-led safeguards over new government requirements.
Six firms sign a short, voluntary pact
The agreement sets out broad commitments rather than detailed public rules for testing or reporting. The Associated Press reported that it calls for advanced AI labs to follow safety protocols and establish internal teams, but that the administration had not provided details on how it would monitor implementation. The pact’s language leaves the participating companies with substantial responsibility for carrying out and assessing their own safeguards.
That arrangement differs from voluntary commitments made by leading AI companies under President Joe Biden, which included specific guidelines for tests and transparency reports, according to Sarah Myers West, a former senior AI adviser to the Federal Trade Commission. Myers West, now co-executive director of the New York-based nonprofit AI Now, said, “Across the board, this self-regulatory approach has consistently done one thing, which is failed to work.”
Companies describe their own checks
Several signatories pointed to steps they say they are already taking. An OpenAI spokesperson told the AP that the company was committed to embedding external assessors and finalizing contracts with third-party safety assessors. Google referred the news agency to a post by CEO Sundar Pichai describing testing and evaluations, as well as work to establish norms and build public confidence.
Meta said it had recently updated standards for training, evaluating and releasing its most capable models. A company spokesperson also pointed to efforts to refine the board’s governance structure. Anthropic told the AP it was bringing in independent evaluators and discussing collaborations with industry groups.
Those company statements describe individual measures, but they do not resolve who will judge whether the safeguards work or what happens if they fall short. The White House reiterated Trump’s comments about balancing innovation and security, but did not directly answer the AP’s questions about how the administration would monitor implementation.
Defense investment draws a different response
As the pact drew attention in Washington, AI founders and investors gathered in San Francisco for Tech Week. At a Wednesday event on investment opportunities, speakers praised the administration’s support for industry. Their focus was on the commercial prospects of AI-powered defense and security systems, not on how the new agreement would be enforced.
Kyle Stanford, a venture capital research director at PitchBook, told the panel that the U.S. government was allocating $1 trillion to defense spending, with $13.5 billion going to autonomous systems and AI. “That has obviously brought a lot of capital into defense tech spending, more than we’ve ever seen,” he said. His remarks offered a view of the economic incentives surrounding AI development as the government places more responsibility for safety on the companies building the systems.
OpenAI’s internal dispute raises the stakes
The pact also arrives amid public disagreement inside the companies expected to police their own work. Three OpenAI researchers said the company had fired them, posting that “fear and unclear rules stymie AI safety work.” David Robinson, who had led part of OpenAI’s safety team before recently resigning, criticized reliance on company-led oversight.
“This is no substitute for hard law, and it does give a lot of freedom to the companies,” Robinson told the AP. He said the time between model generations is measured in weeks and warned of serious near-term safety challenges. That concern sits uneasily beside an agreement that asks labs to build internal monitoring teams without publicly spelling out a government process for checking their work.
The dispute leaves the pact’s central promise—companies monitoring their own AI development—dependent on internal teams, company-specific evaluations and whatever external assessments each firm adopts. The White House has yet to detail how it will monitor the six signatories’ implementation.