Antitrust Law Daily Wrap Up, ANTITRUST NEWS: Criminal enforcement ADAAG Daniel Glad warns Antitrust West Coast Conference attendees about algorithms, AI, (May 15, 2026)

By Justin Marcus Smith, J.D.

Glad said concern about algorithmic coordination has never been about “the math,” it is about the substitution of shared non-public competitive information for the independent decision-making required of competitors.

Acting Deputy Assistant Attorney General Daniel Glad, of the Department of Justice Antitrust Division (AD), addressed the use of algorithms and AI large language models (LLMs) in remarks before the Antitrust West Coast Conference on May 14, 2026. Glad presently leads all criminal antitrust enforcement in the U.S. Glad warned that although collusive tools and cover stories may change, core criminal conspiratorial conduct does not, and the use of technology only makes it more detectable than ever. Even so, in Glad’s view, Acting Assistant Attorney General Omeed Assefi recently gave an accurate description that the AD is striving to balance a strong American AI industry with vigilant antitrust enforcement. Glad addressed three main topics: (1) antitrust doctrine in the context of advanced software use; (2) AD detection and enforcement tools already on-hand; and (3) the questions the AD will pose in the context of algorithmic and LLM use. In Glad’s view, the use of software only makes collusion more detectable, and excuses like “we just follow the model” or “the model did it” will beg the question of intent.

Antitrust doctrine. Glad began with the observation that the most developed area of algorithmic antitrust enforcement involves software platforms that “aggregate competitor and return pricing recommendations to those same competitors.” He said RealPage is the paradigm case. RealPage ultimately agreed to:

• use only historical rental data aged at least twelve months;

• not report rental pricing more narrowly than statewide;

• submit to court appointed monitor;

• operate under written antitrust compliance policy approved by Antitrust Division; and

• cooperate in U.S. continuing civil case against landlord defendants.

Glad summarized that the RealPage consent judgment was a precise civil remedy, aimed at the specific mechanics that turned a pricing tool into an information exchange, that proscribes real-time inputs and granular outputs. The remedy mapped to the harm. It did not ban the software or algorithmic pricing generally. The AD is not concerned with the mathematics used. The concern about algorithmic coordination is about substitution of shared non-public competitive information for the independent decision-making legally required of competitors.

Glad’s second observation was that even though the RealPage settlement involved a civil remedy, criminal charges “are on the table” if there is evidence of an illegal agreement to coordinate. Glad warned the AD will treat any system that replaces independent decision-making with shared competitive intelligence for what it is.

Glad’s third observation was that the courts are already sorting through algorithmic antitrust doctrine on a parallel track. For example, a district court in Seattle denied a motion to dismiss in 2024 holding that algorithmic price-fixing allegations could state a per se antitrust claim under Section 1 of the Sherman Act. Other courts that reached the opposite conclusion did so because there was no pleading of a rim agreement among competitors. The common question at the dismissal stage in such cases is whether there is an agreement among competitors. That determines whether pleaded conduct is a vertical arrangement subject to rule of reason or a horizontal conspiracy subject to the per se rule.

Detection tools. Glad noted next how the Antitrust Division has built its investigative architecture over many years. He said the essential building blocks of cartel detection are durable and transferable to modern cartel conduct, across media, including the use of software.

He said the Procurement Collusion Strike Force (PCSF) is the leading example of transcendent detection. The involvement of software only makes detection more effective. Glad said the AD has trained more than 47,000 federal agents and compliance professionals to detect procurement collusion, and defense counselors who work in this space take the custodial sentences seriously. The same software automation and corresponding detection logic applies and is readily transferrable to the commercial markets where it might be used for horizontal coordination. When competitors coordinate through a software-as-a-services (SaaS) tool, for whatever reason, the “digital residue” is all the more extensive. Glad noted how the PCSF has invested in the infrastructure to use that residue with the Data Analytics Project, an interagency collaboration on red-flag detection in procurement data.

The Whistleblower Rewards Program is “another piece of the investigative architecture” and the most recent addition to the AD detection and enforcement toolkit. The interplay between the Whistleblower Rewards Program and the AD’s longstanding Leniency Program has changed the incentive structure by creating a second race to not only be the first company to report, but to be the first company to report before an insider does. The addition of the insider dynamic is key in the algorithmic context because algorithmic arrangements tend to be visible to a much wider group, including software engineers and data scientists, product managers who sold the software, and compliance professionals who review it. Although the Whistleblower Rewards Program was not designed with AI in mind, Glad said it is well-positioned to uncover algorithmic anticompetitive behavior.

Glad added that the advent of the Whistleblower Rewards Program means general counsel must be aware that a company’s path to Leniency may close while a company is deliberating about whether to self-report. At the same time, the AD has made an effort to make self-reporting easier and clearer. Although the new Corporate Enforcement Policy does not apply to criminal violations of the Sherman Act, it leaves the Leniency Policy in place and even provides a flowchart so self-reporting companies can “come in and resolve all their criminal exposure.”

Questions to expect. Glad said the first antitrust question to expect about LLM use moving forward will be what the models do with non-public data involving competitors. Simply put, if a pricing system depends on competitor confidential inputs to function, in-house counsel should expect the AD to ask why that is not anticompetitive coordination. The AD will apply hub-and-spoke analysis and ask whether there was a meeting of the minds. Training inputs could make a model a collusive hub for anticompetitive spokes. The AD expects this to be the very purpose behind the creation of some AI models.

Next, the AD will look at intent or mental state based on what conspirators said, wrote, and did. Glad noted that intent travels with the human decision to contribute to and rely on the system. It does not turn on who coded it. For those who lead enterprises, intent does not turn on the use of some tool or agent to carry out a scheme, but whether those involved knowingly used a tool to do what could not be done lawfully in a direct manner.

The third question is whether the per se rule is applicable. Glad said the courts are already doing the work of sorting hub-and-spoke algorithmic arrangements from ordinary vertical software contracts. Pricing by LLM will be what the courts tackle next. The AD view is that classification does not change because software is newer, it tracks the conduct. The per se rule will apply whenever competitors have agreed, whether through information sharing or through follow-the-algorithm understandings, to eliminate competition. Glad concluded here that the AD thinks about the three questions he outlined here “every single day.”

Compliance. With respect to compliance and risk mitigation in the algorithmic era, Glad recommended that in-house counsel read the risk-assessment questions specifically directed at artificial intelligence and algorithmic tools, as included in the November 2024 update to the AD Evaluation of Corporate Compliance Programs in Criminal Antitrust Investigations. In-house counsel need to know which pricing and procurement tools are in use, the data inputs, the data outputs, and how any pooling of non-public competitor data relates to the antitrust laws. In-house counsel also need to train sales, pricing, and procurement teams on the guardrails of conduct not only around the tools “but also around the conversations that happen among the users of those tools.” The compliance function must anticipate the implications of repeated interactions with competitors, trade association participation, what employees are discussing, data flows, and whether the exchange of competitively sensitive information through third-parties is legitimate. In-house counsel must place antitrust questions on par with the privacy, cybersecurity, and other risks of using AI.

Glad warned that “we just follow the model” or “the model did it” is not compliance and only begs the question of intent. Moreover, advice of counsel is not a cognizable defense to a Sherman Act violation.

He also warned the move from human conversations to machine-assisted ones does not change who is accountable. The AD charges individuals who may serve sentences. Personal exposure can extend to engineers, data scientists, product managers, account managers, and compliance professionals. He said the AD will look for any conduct that replaces competition with collusion.

MainStory: TopStory Antitrust AntitrustDivisionNews AINews GCNNews