Former Antitrust Chief Says AI Labs Don't Need an Exemption to Build Safely
Former antitrust chief Jonathan Kanter says leading AI companies do not need an antitrust exemption to build safe products, in a Decoder interview published Sept. 19, 2026 as the first part of a two-part series on the future of business. Decoder
Kanter was the antitrust chief for the U.S. Department of Justice in the Biden administration. He is now a professor of technology policy at Carnegie Mellon.
While leading the Antitrust Division, he brought cases against Google, Apple and Ticketmaster. The case against Apple remains live under the Trump DOJ.
The background here matters because enforcement actions often stop when administrations change. This one did not. The Apple litigation runs from a Biden-era filing to Trump-era prosecution, and that continuity gives Kanter's current comments weight beyond hindsight.
The Decoder interview focuses on competition in artificial intelligence. CEOs of major AI companies have asked for antitrust exemptions, or legal permission to coordinate without fear of lawsuits, to work together on safety issues. Kanter rejects that idea.
He made the same point days earlier on television. Kanter joined CNBC's 'The Exchange' to discuss the future of AI regulation. His position was that AI companies do not need antitrust exemptions to deliver safe products. CNBC
That position was echoed by former Trump AI czar David Sacks, who retweeted Lina Khan saying there is no need for an antitrust exemption.
Kanter's past work covers enforcement across technology, healthcare, housing, media, labor and defense as U.S. Assistant Attorney General. He has previously said the antitrust fight against Big Tech is just beginning. He issued a statement regarding U.S. antitrust enforcers' plans to block Big Tech mergers. On Apple specifically, he said the company's choices had made its system "less private and less secure." The Verge
Since leaving the DOJ, Kanter has stayed in the policy conversation. He took part in a fireside chat with Congressman Chris Deluzio hosted by Carnegie Mellon University's Block Center. Carnegie Mellon
The broader context here is the old tension between coordination and collusion. Safety cooperation can mean joint testing protocols, shared incident data or common evaluation benchmarks, which are standard tests to compare models. Competition lawyers worry the same request can lead to joint standard-setting that hardens into market allocation, or safety talks that drift into model release timing and pricing.
Looking at what this means for builders, the exemption debate is less about whether labs can build safely and more about what kind of teamwork gets legal protection. Antitrust law already permits teamwork that helps competition. It blocks teamwork that restrains trade or protects incumbents. For engineers and product leaders, that line affects API access, promises about interoperability, or systems working together, and whether evaluation tools stay open to smaller entrants.
In my view, Kanter's intervention is useful because it narrows the question. If safe products do not require an exemption, then the request for one is really a request for a different way to handle risk. It would shift oversight from outside review toward agreement among peers. The industry has tried peer agreement before, in privacy, in content moderation and in app distribution. The results were mixed. Independent enforcement, however slow, kept the boundaries visible.
Worth flagging for the longer arc, competition tends to expand who gets to build. Open interfaces, second-source suppliers and viable challengers have done more for adoption than any single safety pledge. I saw this with my own kids as smartphones, cloud services and then generative models each moved from novelty to utility. Each shift looked centralized at first. Each opened up as tooling, talent and infrastructure spread. Keeping that path open in AI will depend less on joint statements and more on whether new entrants can obtain compute, distribution and customers on workable terms.
