The White House has secured voluntary commitments from six leading artificial intelligence companies to allow independent auditors to assess their most powerful AI models for potential risks, including those related to cybersecurity and biological weapons. The agreement, which involves OpenAI, Google and Meta among others, does not impose penalties for non-compliance or set a deadline for implementing the checks.
The pact represents a continuation of the administration's strategy of seeking voluntary industry safeguards rather than pursuing binding legislation. Previous rounds of White House commitments have included pledges on watermarking AI-generated content and sharing safety test results with the government.
Under the new agreement, companies have pledged to give external auditors access to their frontier AI systems before and after public release. The audits are intended to evaluate whether models could be used to carry out harmful activities such as launching cyberattacks or creating biological agents.
Critics have questioned the effectiveness of a purely voluntary framework, noting the absence of any legal requirement or deadline. Without mandatory compliance, some experts argue that companies may delay or water down the audits. Supporters of the approach say it allows for rapid adaptation as the technology evolves and avoids the pitfalls of rigid regulation.
The six companies have not specified which external auditors will conduct the reviews or how often they will occur. The White House has indicated it expects the audits to begin within months, but no firm date has been set.
This deal comes amid growing global concern about the pace of AI development. Several countries, including members of the European Union, are moving toward binding regulations, while the U.S. Congress has yet to pass comprehensive AI legislation.