Joe Biden moves to compel tech groups to share AI safety test results

Stay informed with free updates

Simply sign up to the Artificial intelligence myFT Digest — delivered directly to your inbox.

Companies whose artificial intelligence models could threaten US national security will have to share how they are ensuring the safety of their tools under a sweeping order by Joe Biden intended to curb the risks posed by the technology.

The order, which the US president issued on Monday, is the broadest step taken by the administration so far in tackling AI threats, from national security to competition and consumer privacy. The measure seeks to mobilise agencies across Washington, including the departments of commerce, energy and homeland security.

“To realise the promise of AI and avoid the risk, we need to govern this technology, there’s no way around it,” Biden told an event at the White House on Monday.

“President Biden is rolling out the strongest set of actions any government in the world has ever taken on AI safety, security and trust,” said Bruce Reed, White House deputy chief of staff. “It’s the next step in an aggressive strategy to do everything on all fronts to harness the benefits of AI and mitigate the risks.”

The order comes as countries across the world grapple with how to regulate AI companies and models that are not naturally captured by individual watchdogs. Gary Gensler, chair of the US Securities and Exchange Commission, told the Financial Times recently that a financial crisis was “nearly unavoidable” within a decade if regulators failed to manage AI risks.

The measures come two days before vice-president Kamala Harris, the Biden administration’s AI tsar, is set to give a speech in London about US policy, before attending the UK’s Bletchley Park summit, where world leaders and tech company executives will discuss potential ground rules for the development of “frontier AI”.

“We intend that the actions we are taking domestically will serve as a model for international action,” Harris told the White House event. She pledged to work with other countries “to apply existing international rules and norms with a purpose to promote global order and stability, and where necessary to build support for additional rules and norms which meet this moment”.

The EU has moved rapidly, drafting tough measures over the use of the technology in a groundbreaking law that is set to be fully approved by the end of the year. But the US is still assessing which aspects of it require new regulation and what is subject to existing statutes.

Asked if the EU law had influenced Biden’s order, a senior administration official said: “I don’t think we’re in a race. I don’t think we’re playing catch-up.” The US had liaised with the EU and a “wide range” of the bloc’s member states about AI regulation, the official added.

Leading industry figures including OpenAI co-founder Sam Altman have toured the globe this year to discuss the potential impact of the tools they are developing. Altman and others have struck a conciliatory tone with regulators but resisted calls to stall or slow the development of increasingly powerful AI.

In May, Altman said his company could “cease operating” in Europe if Brussels’ efforts to regulate the technology were overly stringent. He later walked back the comments.

Biden’s order escalates US AI policy after 15 companies — including Amazon, Google, Meta, Microsoft and OpenAI — made voluntary commitments earlier this year to manage the technology’s risks.

The White House will use the Defense Production Act, a cold war law used at the peak of the Covid-19 pandemic, to compel businesses developing AI models posing serious risks to national security, economic security, or public health to notify the government when training these systems and to share their safety test results.

To date, companies vying for AI dominance have typically preferred to keep the designs of their models private.

“If organisations don’t adhere to that law, we could bring Department of Justice actions in appropriate manner to enforce that,” said the senior official. But he stressed that these requirements would “primarily” capture the next generation of the world’s most powerful AI tools and would not “catch any system currently on the market”.

Under the order, the commerce department must craft guidance on adding watermarks to AI-generated content in a bid to tackle “fraud and deception”, including deepfakes.

The measures also seek to promote competition in the AI sector and encourage the Federal Trade Commission “to exercise its authorities” at a time when US antitrust regulators have warned against potential monopolies arising from the technology’s structural dependence on scale.

The order addresses privacy risks, urging Congress to pass data privacy legislation while seeking an assessment of how agencies collect and use “commercially available information”. It also calls for measures to curb harms caused by AI on workers and medical patients as well as to address “algorithmic discrimination” in housing, healthcare and justice.

The extent to which the order will be implemented as intended remains unclear. While the White House can avail itself of certain laws to bring enforcement actions or direct departments to develop guidelines, it may only encourage other independent agencies, such as the FTC, to implement Biden’s plan.

The senior official said that while executive orders “have the force of law”, Biden has said “that we were going to need bipartisan legislation to do more in artificial intelligence”. Many of the president’s priorities “require legislative action to fully execute . . . We are not at all suggesting this is the end of the road on AI governance.”