White House assigns tech giants lead on AI safety

The White House announced a voluntary agreement that places primary responsibility for testing and reporting AI safety with major technology companies, including Microsoft, Google, OpenAI, Amazon and Meta. The pact asks firms to run safety testing, share incident reports and adopt common safeguards for large-scale models before wide deployment.

Under the agreement, participating companies will carry out internal risk assessments, conduct adversarial testing and develop mitigation measures prior to public release of powerful systems. The administration described the accord as a way to speed deployment of new tools while reducing chances of harmful outcomes and assigned firms the lead role in identifying technical risks such as misinformation, model misuse and safety failures.

Officials framed the pact as a set of baseline expectations rather than binding law. Companies are expected to publish regular transparency reports, notify federal agencies about major security or safety incidents and permit independent third-party audits of selected systems. The framework encourages access controls for higher-risk models, logging of key development decisions and common standards for stress testing model behavior.

The administration indicated firms that meet the agreement’s standards could receive regulatory flexibility. Companies that fail to follow the framework may face increased oversight. An interagency committee will monitor compliance and work with industry on technical specifications. The White House described the arrangement as temporary while Congress considers formal legislation.

Industry executives welcomed the agreement as a way to coordinate safety work across competing firms and to avoid fragmented rules that could slow product rollouts. Company statements emphasized ongoing investment in red-teaming, content moderation tools and internal review processes, and pledged to share certain safety research and techniques through neutral forums.

Civil society groups and some lawmakers expressed reservations about leaving primary oversight to the companies that build the systems, noting that voluntary commitments lack clear enforcement. Members of both parties continued to push for statutory rules and clearer liability standards for high-risk AI systems.

The pact follows earlier U.S. guidance and executive actions aimed at promoting risk assessment and model transparency. International regulators have moved toward binding rules, and U.S. officials said the agreement is intended to complement, not replace, upcoming legislative work.

Next steps require companies to publish implementation plans and the interagency panel to issue technical guidance on testing protocols and reporting thresholds. Observers identified key measures of success: how quickly firms adopt shared standards, the transparency of harm reporting and whether the government issues enforceable rules if private practices fall short.