Jonathan Kanter has discovered the virtues of capitalism. As head of the Justice Department’s Antitrust Division under President Biden, he sued Google and Apple and turned the division’s attention to concentration in artificial intelligence. Now he is an AI entrepreneur himself.
On October 28, Kanter and Doha Mekki, his principal deputy and later the division’s acting chief, launched FluxCap, which they bill as a “Deep Legal Tech” platform. It arrives with $6.4 million in seed money from Y Combinator, Bloomberg Beta, Omidyar Network, the litigation funder Burford Capital and others. The founders promise that small teams of lawyers will be able to “litigate at scale” against opponents with far deeper pockets.
One almost expects them to invoke the consumer welfare standard.
The irony is real, but it is not the most interesting thing about FluxCap. The company is a test of a question antitrust has argued about for half a century: what counts as harm to competition? Software that builds cases faster will build whatever cases its designers think are worth building. So the standard it applies matters more than how fast it runs.
The pitch
FluxCap’s founders start from a real problem. Kanter told ABC News that his team at Justice could pursue only about 10 percent of the cases it thought worth bringing. “We were completely outnumbered,” he said. A major monopolization case means years of document review, depositions and economic analysis, and the other side can usually afford more of all of it.
FluxCap says its software reviews the evidence, finds the gaps in it, tests a lawyer’s assumptions and anticipates the defense’s arguments, with every fact and citation open to audit. The founders insist the tool supplements lawyers rather than replacing them. “We’re not trying to be a better lawyer,” Kanter said; “we’re trying to help lawyers be better.”
The customers are clear enough. The company describes its mission as serving “public and private plaintiffs,” and press coverage describes the target market as plaintiffs’ lawyers and government enforcers. In-house legal departments are on the customer list too, but the pitch is written for the side that files the complaint, not the side that answers it.
If the software works as advertised, it lowers the cost of producing an antitrust complaint. That is good economics. Lower costs, greater output and easier entry for small rivals are exactly what antitrust law exists to protect, and it is gratifying to see the founders put those forces to work.
The economics of cheaper lawsuits
Litigation responds to prices like anything else. Make complaints cheaper to produce and more of them will be filed. Some will be strong. Others will survive mainly because fighting them costs more than settling them.
Antitrust is unusually exposed to that second kind. Section 4 of the Clayton Act gives a successful private plaintiff three times its damages plus a reasonable attorney’s fee. A defendant that wins recovers neither. Trebling was meant to encourage private enforcement, and it does. It also means that even a modest chance of losing carries a large expected cost, which is why so many antitrust cases settle before anyone decides whether the conduct actually hurt consumers.
A defendant pays for discovery, outside counsel and executive time whether or not it broke the law, and a long public fight carries reputational costs no verdict repairs. Software that cuts what it costs a plaintiff to build a case, without cutting what it costs a defendant to answer one, widens a gap that already pushes companies to settle weak claims.
The obvious reply is that defendants will buy AI too, and large defense firms are adopting it quickly. If both sides’ costs fall, perhaps the balance holds. But the savings are not symmetric. AI is very good at reading documents and drafting arguments, which is most of what it takes to file a complaint. It does far less about the costs that fall on the defendant alone: producing millions of pages in discovery, sitting executives for depositions, and living with an open accusation for years. Cheaper drafting helps the side that drafts first.
Burford’s place among FluxCap’s investors sharpens the point. Burford is the world’s largest litigation financier; its business is paying for lawsuits in exchange for a share of the recovery. Nothing public suggests Burford has agreed to finance cases built on FluxCap. But the combination is easy to imagine. AI lowers what a case costs to build, outside money removes the risk of losing it, and treble damages raise the payoff of winning. That is the recipe for an attractive asset class, whether or not the underlying conduct harmed anyone.
Competitors versus competition
The more important question is what the software believes an antitrust violation is.
Antitrust has never been short of clever theories. Its hard problem, for more than a century, has been telling conduct that injures competitors from conduct that injures competition. A firm that cuts prices, builds a better product or tightens its supply chain will take customers from its rivals. That is competition working. The rivals will still call it exclusion, and they will have the documents to show they lost business.
The Supreme Court got this wrong for much of the 1960s. In Brown Shoe (1962), the Court declared that the merger law protects competition, not competitors, and then blocked a merger anyway, in part to preserve small, locally owned businesses. It said plainly that Congress had accepted “occasional higher costs and prices” as the price of that protection. In Von’s Grocery (1966), it stopped a merger of two Los Angeles grocery chains with a combined share of about 7.5 percent, prompting Justice Stewart’s famous dissent that the only consistency in such cases was that “the Government always wins.”
Utah Pie (1967) is the purest example. National bakers cut prices on frozen pies in Salt Lake City, where a local firm dominated the market. Utah Pie’s share fell, but its sales kept growing and it stayed profitable. Consumers got cheaper pies. The Court nonetheless held that the evidence could support a jury verdict against the national firms, pointing to the market’s “drastically declining price structure” as evidence of injury. Falling prices, in other words, were treated as the harm.
Those decisions are what The Antitrust Paradox was written against. Over the following decades the courts came around to the view that antitrust should ask whether conduct leaves consumers worse off, and later cases such as Brooke Group (1993) made predatory-pricing claims like Utah Pie’s very hard to win. But the old instinct never disappeared. Every rival that loses business has a story about exclusion, and every one of those stories can be written up as a complaint.
Kanter’s own docket shows that the question is still live. In 2024 his division sued Apple, arguing that its restrictions on apps, messaging and smartwatches unlawfully protect the iPhone. Apple answers that those same restrictions deliver the security and seamless integration its customers pay a premium for. Proving harm to rivals is the easy part of a case like that. The hard part is whether consumers are worse off, and no amount of document review settles it.
What a good tool would do
This is why the standard a tool is built on matters more than its speed. An AI that reads ten thousand internal emails will find a great deal of evidence that a company wanted to beat its rivals. Every successful company wants that. Evidence of intent to win is cheap and abundant. Evidence that consumers paid more, got less or lost something better is scarce, and it has to be argued from prices, output and quality, not from the tone of an executive’s memo.
A system built to ask the consumer-welfare questions first would be a real contribution. Did prices rise? Did output fall? Did quality or innovation suffer, compared with what the market would otherwise have produced? Software that put those questions at the front of every analysis, and told its user honestly when the answers were missing, would make enforcers more selective rather than merely more prolific. It would screen out weak cases as efficiently as it builds strong ones. Government enforcers working on a fixed budget should want exactly that.
A system built on the more expansive theories of harm Kanter pressed in office would do something else. It would turn out more polished versions of the cases the consumer welfare standard was designed to screen out, and it would turn them out at scale.
FluxCap says its system tests assumptions and anticipates the other side’s arguments, and nothing yet suggests it favors unsound theories. That is to its credit. But assumptions are tested against a standard, and the founders have not said which one theirs applies. It is a fair question to put to the two people who led an Antitrust Division that argued the prevailing standard was too narrow.
The paradox, again
Kanter and Mekki are betting that a better, cheaper product will take business from established law firms and legal-tech incumbents that cannot match it. If they succeed, some of those rivals will lose clients, revenue and perhaps their independence. No one will call it exclusionary. They will call it innovation, and they will be right.
That is the lesson of their own venture: innovation, scale and lower costs are how competition works, and the rivals left behind are not evidence that it failed. It would be encouraging if their artificial intelligence applied that insight as enthusiastically to the businesses it investigates as its founders have applied it to their own.