Washington DC - President Donald Trump signed an executive order Tuesday creating a voluntary framework under which AI developers will share advanced models with the government before public release.
The central provision allows companies such as OpenAI, Google, or Anthropic to give the government access to their most powerful models for up to 30 days before planned release.
The order was triggered by concerns over Anthropic's Mythos model, which the AI startup has held back from the public due to its ability to expose vulnerabilities in computer systems, including those of banks, governments, and hospitals.
The 30-day window represents a compromise. The original draft called for up to 90 days of pre-release government access, while tech companies had pushed to cut that figure to just 14 days.
Kent Walker, Google's head of public affairs, called the order an "important step forward" that will ensure "defenders have the AI tools they need to keep America secure."
The signing comes after a turbulent few weeks in which the White House appeared close to unveiling the measure, only to pull back abruptly.
AI safety advocates warn new order is still "not enough"
According to Politico and other media, David Sacks, the Silicon Valley venture capitalist who served as Trump's AI and crypto czar, called the president to warn that the measure would slow innovation and hurt the US in its AI race with China – blindsiding White House staff who believed Sacks supported the order.
Sacks wrote on X last week that "unnecessary regulation is the biggest threat to innovation in America," adding that winning the AI race required clearing "bureaucratic hurdles" from state legislatures and "woke" Washington politicians.
Trump scrapped a Biden-era AI oversight order on his first day back in the White House.
Biden's 2023 order required AI companies to share safety test results with the government and leaned heavily on voluntary commitments – already a light-touch approach that fell short of what many experts had called for.
By contrast, the European Union's AI Act – which entered into force in 2024 – sets binding rules for high-risk AI systems, including mandatory transparency requirements and, for the most powerful models, obligations around safety testing and incident reporting.
"This is an important step in the right direction," said Anthony Aguirre, CEO of Future of Life Institute, which advocates for AI safety.
"Voluntary frameworks are not enough, however," and the government must be empowered "to block the release of systems that pose an unacceptable national security risk," he added.