Trump secures voluntary AI safety accord with tech leaders

US President Donald Trump has reached a voluntary safety accord with six leading artificial intelligence companies, mandating internal controls, independent audits, and board oversight for their most advanced systems. The agreement, signed following a White House meeting on Tuesday, includes OpenAI, Anthropic, Google, Meta, Nvidia, and Elon Musk's xAI. Trump characterized the commitment as "morally binding," though the document itself does not enforce compliance or impose penalties.

The document requires these companies to implement four layers of controls and audits. The initial layer mandates internal safeguards to monitor advanced systems during development, addressing concerns related to cybersecurity, biosecurity, chemical threats, and unintended AI model actions. Each company must also establish an internal team to verify the effectiveness of these safeguards.

A subsequent layer involves external auditors or evaluators independently assessing the established controls. The final layer entails an independent committee from each company's board reviewing reports from internal teams and auditors. Despite retaining responsibility for developing their technology safely, the companies will collaborate to establish standards and best practices.

Trump emphasized that the accord would involve "tremendous self-policing" and compared it to a constitution, acknowledging the potential consequences of AI systems causing harm. The administration is considering a committee of 10 individuals to oversee the initiative. Although the accord introduces an external-review requirement, it does not necessitate the publication of audit findings or establish penalties for deficiencies.

This arrangement has sparked debate among safety researchers and advocates, who argue that voluntary commitments may lack accountability due to companies' discretion in implementation. Conversely, proponents of industry-led standards argue that fast-evolving technology renders rigid rules obsolete and that specialized companies are best positioned to identify emerging technical risks.

The White House meeting brought together executives from OpenAI, Anthropic, Google, Meta, Nvidia, and Musk. Microsoft's Satya Nadella and other technology leaders also attended the broader gathering. During the event, Anthropic's Dario Amodei highlighted the potential risks associated with increasingly powerful AI systems. The accord brings together companies with divergent public stances on regulation while avoiding mandatory standards that some industry figures and lawmakers have sought.

The development of autonomous AI agents poses significant concerns, with models now capable of discovering software vulnerabilities, executing complex cyber operations, and assisting in scientific tasks that could pose biological or chemical risks. Public opinion, as reflected in a recent Reuters/Ipsos poll, shows that 73% of respondents believe AI companies have not done enough to ensure safety, increasing pressure on Washington and the industry to demonstrate stronger safeguards.