President Donald Trump and Big Tech executives signed a voluntary pact on Tuesday aimed at self-policing artificial intelligence (AI) through a four-layered safety framework even as the administration maintains the U.S. cannot slow development in its global competition with China.
The one-page, 300-word “White House Accord on Super Intelligence: Joint Commitment on Frontier Responsibilities” was posted to Truth Social following a White House meeting with tech leaders. Signatories included Google CEO Sundar Pichai, Anthropic CEO Dario Amodei, Meta Platforms Inc. CEO Mark Zuckerberg, OpenAI President Greg Brockman, xAI founder Elon Musk, and NVIDIA Corp. CEO Jensen Huang.
Speaking to reporters after the gathering, Trump praised the initiative, characterizing the voluntary agreement as “morally binding” and describing it as “almost like a constitution” for the sector.
“The biggest people in the world signed that, and I signed it as president, and it really is a form of protection,” Trump said, arguing that the industry should largely regulate itself. Asked by a reporter whether guardrails beyond good intentions were necessary, Trump downplayed concerns about bad actors, emphasizing that tech leaders “love our country and the world.”
The framework outlines four principal layers of internal and external oversight:
Robust internal controls. Companies commit to monitoring model capabilities and alignment during training and deployment, specifically targeting risks in cybersecurity, biosecurity, and chemical threats to prevent unintended system access or hacking.
Internal oversight teams. Dedicated internal groups will verify that monitoring and detection mechanisms operate as intended and resolve identified issues.
Independent external reviews. Companies will partner with third-party auditors or evaluators to independently assess system controls.
Board-level oversight. Participating firms will establish independent board committees to review safety reports from both internal teams and external auditors.
While the accord notes that “over time, it may make sense to codify these steps into laws or regulations,” the agreement currently carries no legal force. Participating companies also agreed to meet regularly to refine safety standards and best practices.
The announcement comes amid mounting friction between corporate leaders and AI safety researchers regarding the rapid pace of development.
Amodei, Musk, and OpenAI CEO Sam Altman have recently voiced support for greater caution, warning that advancing capabilities could outstrip existing safety protocols. Anthropic researchers have publicly cautioned about catastrophic risks, including potential threats to critical infrastructure or biosecurity, if models are deployed without adequate alignment safeguards.
Meanwhile, Microsoft Corp. co-founder Bill Gates called AI “the most dangerous thing humans have ever gone near,” and argued that existing government frameworks are ill-equipped to handle AI’s rapid capabilities. He highlighted immediate threats from bad actors using the technology to bypass specialized expertise, noting AI can now enable bioterrorists to launch mass-fatality attacks or trigger cyberstrikes capable of crippling power grids and financial systems. Gates emphasized that regulators must act quickly to manage both current malicious uses and future control risks.
Despite the warnings, Trump reiterated his resistance to federal intervention that could stall American innovation. Emphasizing the geopolitical race with China, Trump stated he would not “stifle” progress.
The White House also signaled a shift in tone by directing executive branch departments to adopt “super intelligence” in place of “artificial intelligence” in official government communications. The directive arrives weeks ahead of midterm elections, where public apprehension regarding rapid technological advances has emerged as a central voter issue.
Industry executives framed the agreement as a constructive step toward transparency. Zuckerberg noted that the framework offers a foundation to give the public and customers confidence that advanced models will perform safely, while emphasizing that the accord serves as an initial starting point rather than a final regulatory solution.
