The artificial intelligence industry recently called for an antitrust exemption to coordinate self-regulation as the risks their models pose to the internet and society grow. Industry-wide self-regulation might well violate antitrust law and would certainly give Big Tech too much power with too little supervision. In the absence of federal action, states can and should provide the required antitrust immunity to enforce safety rules so long as they oversee the conduct, write Noah Hoffman and Fiona Scott Morton.
On September 12, Anthropic co-founder and chief executive Dario Amodei set off a firestorm by calling for artificial intelligence companies (including his own) to slow down the development of their models. He wrote that AI models will soon “outrun our ability to understand and control” them and warned that without a concerted effort to slow the pace of development, the models could soon “be capable of taking over the entire internet.”
Amodei promulgated a three-step plan to mitigate this risk and called for AI companies in the United States to “voluntarily work together to set standards” to keep “capabilities in balance with safety.” Sam Altman (CEO of OpenAI), Elon Musk (owner of xAI), and Demis Hassabis (chief of Google DeepMind) quickly endorsed Amodei’s proposal.
Due to the collaborative nature of the plan, Amodei noted that it would be “legally challenging” “[f]or antitrust reasons.” He is correct that some forms of industry-wide self-regulation would violate the antitrust laws. U.S. Supreme Court precedent holds that a simple agreement among competitors to reduce product quality is a per se violation of the Sherman Antitrust Act whether the agreement is made through an industry trade group or written by its largest players. Because of the legal challenges, Amodei called for Big Tech to be granted a “narrow waiver” from federal antitrust laws to “coordinate” safety standards and the rate of AI progress.
The reaction from outside the industry to Amodei’s proposal was immediate and negative. Politicians shut down the idea. Competition experts balked, saying an antitrust waiver would “allow the leading AI labs to entrench their positions by setting the rules of the game and determining who gets to compete.” Further, populists on both the left and right do not think Big Tech cares about the welfare of humanity nor do they trust big business to regulate itself. The standard way to deal with a safety or quality problem like this is for the government to regulate, rather than letting businesses decide the rules.
But the problem is that the federal government is not regulating AI and shows no signs of doing so. In an unusual twist, citizens and corporations are on the same side—both want regulation. The federal government is failing to deliver.
There is a solution: states, without any action from Congress or the White House, can provide AI companies with immunity from federal antitrust laws. All a state needs to do is articulate its policy interest (such as protecting its citizens from mass annihilation) and develop a scheme to oversee the conduct. Companies cooperating with each other under such a scheme cannot be held liable under federal antitrust laws.
The relevant doctrine is called state-action immunity. As the name suggests, it provides immunity to private actors from federal antitrust laws, so long as the state’s policy of displacing competition is “clearly articulated and affirmatively expressed” and the private conduct is “actively supervised by the state.” The doctrine, sometimes called Parker immunity, was first established by the Supreme Court in the 1943 case Parker v. Brown. It has been expanded and reaffirmed many times since, most recently in 2015.
State-action immunity has been found to protect transportation companies collaborating to set shipping rates, a municipal contractor monopolizing parking meters, and an advertising company protecting its billboard monopoly by restricting new development. The doctrine has also been used to immunize anticompetitive hospital mergers when state politicians believe the combined firm will advance other state interests such as retaining rural facilities or protecting union jobs.
States such as California, Texas, and others should immediately enact state-action immunity laws to enable a supervised AI slowdown. The legislation must explain that the state’s intention is to displace competition with supervision for AI product safety. It also must direct the state attorney general to supervise the industry’s collaboration to ensure it advances safety without harming other dimensions of competition. Under current doctrine, only one state needs to supervise the collaboration to grant it immunity, so long as the economic effects of the collaboration are significant in that state. If multiple states pass state-action immunity laws, they can supervise the collaboration in concert. Doing so may make the immunity more robust to legal challenges.
We recently explained this idea in the context of both companies and citizens wanting collaborations among private companies to address climate change. The same analysis applies to collaborations among AI companies.
The AI industry says we are short on time before its models become truly dangerous. A state law conferring immunity could be drafted in a matter of days because it would not require the state legislature to devise regulations or compromise on policy. The law need only articulate the state’s intention to displace competition with supervision and direct the state attorney general to implement the law.
Once a law is passed, the state attorney general can promulgate procedures AI companies must abide by to receive immunity. The procedures should safeguard as much competition as possible. For instance, the attorney general should require that all meetings between competitors include representatives from his or her office. The attorney general should also require access to every document exchanged between competing companies, and all exchanges of competitively sensitive information should be pre-approved. Most importantly, the attorney general must approve all the companies’ agreements and jointly adopted procedures to slow down development of AI or otherwise restrict independent decision-making. The attorney general should only approve agreements and procedures that advance the state’s goals of increasing AI safety while preserving as much competition as possible.
State-action immunity for the AI industry benefits both the companies and the public. Companies obtain immunity from federal antitrust laws and can propose safeguards they believe will be effective to protect us all. The entire scheme is voluntary to enter, but compliance is not voluntary after the companies commit to certain behaviors. The public is ensured that any self-regulation or agreement between competitors advances AI product safety without limiting competition in other aspects of the companies’ businesses. State supervisors will be in the room, overseeing and approving every aspect of the collaboration. Best of all, no federal legislation is required to implement the scheme.
Authors’ disclosures: Fiona Scott Morton is an economic expert in the United Kingdom for a group of advertisers seeking damages from Google, and in other competition cases both in the United States and United Kingdom. She regularly works as an expert witness for government plaintiffs on matters that are confidential. Noah Hoffman reports no conflicts of interest. You can read our disclosure policy here.
Articles represent the opinions of their writers, not necessarily those of the University of Chicago, the Booth School of Business, or its faculty.
Subscribe here for ProMarket’s weekly newsletter, Special Interest, to stay up to date on ProMarket’s coverage of the political economy and other content from the Stigler Center.
