Biden issues sweeping executive order on frontier AI safety

For the past year, leading AI labs have essentially graded their own homework before releasing massive models. That free pass is officially over.

President Joe Biden signed a sweeping Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence. Marking the White House's first direct executive order dedicated to AI, the directive establishes sweeping federal guardrails across national security, privacy, biological risks, and algorithmic bias.

"There’s no other way around it," Biden said at the signing ceremony. "In my view, it must be governed."

The Defense Production Act gets an AI upgrade

The most aggressive move in the order invokes the Defense Production Act, a Korean War-era law typically reserved for military supply emergencies.

Under this authority, companies training foundational models that pose potential risks to national security, economic security, or public health must notify the federal government during the training phase. Even more demanding: developers must hand over the results of all internal "red-team" safety testing to the government before making those models public.

The National Institute of Standards and Technology (NIST) will draft standard benchmarks for that red-team testing. Meanwhile, the Department of Homeland Security (DHS) will establish a new AI Safety and Security Board to protect critical infrastructure against cyber, chemical, and biological threats potentially amplified by AI.

To combat deepfakes and AI-driven fraud, the Department of Commerce is tasked with developing standards for authenticating official communications and watermarking synthetic media. Federal agencies will be required to adopt these watermarking tools first, aiming to establish an authentic baseline for official communications.

Deadlines for government contractors and the workforce

The executive order also lays down concrete timelines for federal agencies and the massive private network that sells to them.

Within 150 days, the Office of Management and Budget (OMB) must issue instructions on how agencies can rely on contractors using federal data for AI development. Within 180 days after that, OMB will release requirements ensuring agency AI contracts protect government data and civil rights. The Department of Health and Human Services (HHS) also faces a 180-day deadline to issue guidance preventing algorithmic discrimination across federally funded healthcare programs.

For employment and labor, the order directs the Secretary of Labor to publish guidance within 365 days to prevent AI hiring discrimination among federal contractors. The White House explicitly warned against using AI for unlawful worker surveillance or degrading job quality, assigning the Labor Department to monitor AI-driven job displacement.

Why it matters

This executive order ends the era of voluntary tech diplomacy. Just months after 15 leading AI companies made voluntary safety pledges at the White House, the administration realized voluntary self-policing was nowhere near enough.

For everyday users and startup builders, open-source models aren't being banned overnight, but the operational climate is shifting fast. If you are developing enterprise software or AI tools for federal agencies, healthcare systems, or critical infrastructure, your compliance burden is about to skyrocket over the next 150 to 180 days as procurement guidelines harden.

We think targeting biosecurity, red-team disclosures, and federal procurement is a shrewd workaround for a divided Congress that cannot pass baseline data privacy laws. However, relying on the Defense Production Act to govern software development is legally unprecedented and could face pushback. Furthermore, executive orders are vulnerable; what one administration mandates, the next can revoke with a single signature. For now, the world's most powerful AI builders must get used to having federal inspectors looking over their shoulders.