Old Crime, New Code: DOJ puts algorithmic collusion in the crosshairs

Software cannot shield collusion. Where it is the path of collusion, criminal enforcement is on the table. That was the message from Acting Deputy Assistant Attorney General Daniel Glad at the Antitrust West Coast Conference on May 14, 2026. In remarks titled “Old Crime, New Code,” Glad delivered the clearest preview yet of how the Antitrust Division will bring criminal enforcement to bear on algorithmic and artificial intelligence (AI)-enabled conduct. The bottom line: you cannot do with the knowing use of an algorithm what you could not do with a simple agreement that does the same thing. When competitors use shared pricing systems to replace independent decision-making with coordinated outputs, criminal charges remain on the table.

What Glad said

Where the criminal line runs

Glad anchored his remarks in the November 2025 RealPage consent judgment, which requires RealPage to use only historical data that is at least 12 months old, limits its reporting to statewide aggregations, and imposes a court-appointed monitor. The judgment does not ban algorithmic pricing, but targets the ingestion of non-public competitor data and the return of granular outputs to those same competitors. Nevertheless, civil resolution in RealPage, Glad cautioned, does not place algorithmic conduct beyond criminal enforcement. Where a horizontal agreement is present and provable, criminal charges remain available.

The key doctrinal question is whether a “rim” (an agreement among competitors) exists. In Duffy v. Yardi Systems, the Western District of Washington allowed per se price-fixing allegations to proceed. In Gibson v. Cendyn Group and Cornish-Adebiyi v. Caesars Entertainment, courts dismissed for failure to plead a horizontal agreement. The takeaway: where competitors knowingly contribute non-public data to a system that shapes each other’s pricing, a rim may exist, and with it, per se liability and criminal exposure.

For companies licensing shared pricing or revenue-management software, the vendor’s intermediary role provides no insulation. The question is whether the company knowingly feeds confidential data into a system whose outputs influence competitors’ decisions. Industries with concentrated competitor sets, such as housing, hospitality, and healthcare, face the highest exposure.

The tools are already built

The Division’s investigative architecture is already built for algorithmic enforcement. The Antitrust Division’s Procurement Collusion Strike Force (PCSF) has trained more than 47,000 federal agents and compliance professionals and secured more than 85 convictions. As procurement migrates to e-platforms and AI-driven bid tools, the digital trail multiplies, which the PCSF’s Data Analytics Project, pre-award data retention, and Inspector General partnerships are positioned to leverage.

The Whistleblower Rewards Program has added a second race alongside corporate leniency. In January 2026, the Division paid its first whistleblower reward of $1 million for information leading to bid-rigging charges in online used-vehicle auctions. Algorithmic arrangements are visible to engineers, data scientists, product managers, and compliance professionals, any of whom could be the next whistleblower. The Department’s March 2026 Corporate Enforcement Policy carves out Sherman Act violations, preserving leniency while offering a complementary path for non-antitrust criminal exposure.

Bid-rigging exposure now extends across the full automation stack. Contractors should expect scrutiny of bid-preparation tools ingesting competitor data, market-intelligence products with opaque provenance, and repeat-bidder communications. The paper trail in algorithmic procurement is broader and more durable than in traditional settings, and with the whistleblower program running parallel to leniency, the window for self-reporting has narrowed.

When the hub is a large language model

Glad previewed three hard questions the Division is wrestling with as AI moves into pricing infrastructure:

• What is the agreement? When competitors feed non-public data into the same model and follow its recommendations, the analysis turns on what the model does with those inputs. Most AI providers train their models with user inputs by default. Where competing users know this, or the terms of service make it such that they should know, the model can become a collusive hub.


• Where does intent lie? Intent, according to Glad’s remarks, “travels with the human decision to contribute to and rely on the system.” It does not turn on who typed the code, and deploying an autonomous AI agent does not insulate the principal who knowingly used it to achieve what could not lawfully be done directly.


• Does the per se rule apply? The classification, says Glad, “does not change because the software is newer.” Where competitors agree to eliminate competition – whether through architecture, information sharing, or follow-the-algorithm understandings – the per se rule applies.

The upshot: where competitor inputs flow into a common model and outputs shape competitor pricing, per se exposure is on the table. “Advice of counsel” is not a defense to a Sherman Act violation, nor is “the model did it” a compliance answer. The Division charges individuals, not just corporations, and the move to machine-assisted coordination does not change personal accountability.

Why this matters

These remarks confirm, publicly and on the record, where the Division is focused. Glad’s remarks were not made on a limb. they are rooted in the Division’s November 2024 Evaluation of Corporate Compliance Programs (Guidelines). Released in the waning weeks of the Biden Administration, the Guidelines reflected policy priorities – including its focus on AI, algorithmic pricing enforcement, and expanded whistleblower protections – that were closely associated with the outgoing administration's antitrust agenda. As new leadership took office in January 2025, commentators and practitioners alike wondered whether the incoming Trump Administration would retain, revise, or shelve the Guidelines altogether. Any doubts about the continued vitality of the Guidelines have been firmly laid to rest with Glad’s speech. Companies using algorithmic pricing, AI-driven procurement tools, or shared revenue-management platforms should treat this as an opportunity to assess their exposure before the Division does it for them.

Compliance in the era of algorithms demands proactive risk assessment

The Guidelines ask whether a company’s risk assessment addresses new technologies, particularly AI and algorithmic revenue-management software; whether the company assesses antitrust risk when new tools are deployed; whether compliance personnel are involved in that deployment; and what steps the company is taking to mitigate risk. Glad echoed these requirements and went further: He envisions robust, proactive risk assessments as a baseline expectation, not just a best practice. Companies are encouraged to consider whether their compliance programs can answer these questions today.

The legal framework hasn’t changed, and that’s good news

Section 1 of the Sherman Act requires an agreement among competitors, general intent, and conduct that falls under the per se rule. Those elements don’t change because the agreement is mediated through software. The rules are knowable, and so are the compliance measures. Companies can assess their algorithmic pricing and procurement tools against established doctrine. The question remains whether competitors knowingly feed confidential data into a system that shapes each other’s pricing. If so, the legal exposure is the same as if they met in person.

The Division’s enforcement tools are already in place

Glad’s speech made clear that the investigative architecture for algorithmic enforcement is not aspirational, it is operational. As stated previously, the PCSF has trained more than 47,000 federal agents and secured more than 85 convictions. The Whistleblower Rewards Program paid its first reward of $1 million in January 2026, and Glad noted that the program is well-positioned for AI-related cases because algorithmic arrangements are visible to engineers, data scientists, product managers, and compliance professionals, any of whom could be the next whistleblower. Leniency used to be a corporate race: first through the door wins. Now there is a second race running in parallel, and the pool of potential whistleblowers in algorithmic contexts is distinctly broad. Companies that take weeks to investigate and deliberate may find that the window for self-reporting has already closed.

Government contractors face heightened scrutiny

Companies holding public procurement contracts, or selling to those that do, should pay particular attention. The PCSF has built a formidable enforcement pipeline: 47,000+ trained federal agents, 85+ convictions, and partnerships with Inspectors General across the federal landscape. The structural features of government procurement (i.e., standardized bidding, documented communications, repeat interactions, institutional record-keeping) make collusion unusually detectable. As procurement migrates to e-platforms and AI-driven bid tools, the digital trail grows. Every automated bid is a structured record; every parameter change is a timestamp. Companies with public contracting exposure should conduct a fresh assessment of their bidding and pricing tools.

Compliance programs will be measured against the Guidelines

The Guidelines remain the operative framework for how prosecutors will assess whether a company has meaningful compliance in place. This is particularly evident given the Final Judgment requirements for RealPage that relate to its antitrust compliance policy and its antitrust compliance officer’s responsibilities. Both requirements echo the spirit and content of the Guidelines. Notably, the Guidelines are prescriptive – for example, they ask what proportion of compliance personnel’s time is dedicated to compliance responsibilities – therefore, companies should evaluate their compliance programs accordingly. In-house counsel should be able to answer these questions today:

• Does the company’s risk assessment address new technologies, particularly AI and algorithmic revenue-management software?


• At what level have C-suite and board members been advised of and approved company policies?


• Does the company assess antitrust risk when new tools are deployed?


• Are compliance personnel involved in that deployment?


• Do compliance training and feedback involve management and the employees who use the new tools?


• What steps is the company taking to mitigate risk?

The Guidelines also ask whether the level of technology devoted to compliance is comparable to the level of technology devoted to other functions. If the answers are unclear, or if no one has asked these questions, now is the time to address that gap.

Individual employees face personal exposure

The Division charges individuals, not just corporations. Sentences in antitrust cases are served by people, not shareholders. The same employees who might serve as witnesses in an algorithmic case (engineers, data scientists, product managers, compliance professionals) are also the employees with potential personal criminal exposure if their conduct crosses the line. Companies should ensure that employees at all levels understand this. “The model did it” and “legal signed off” will not be defenses.

The case law is actively developing

Courts are still sorting out when algorithmic pricing arrangements cross the line into per se violations. Duffy v. Yardi Systems allowed per se algorithmic price-fixing allegations to proceed; Gibson v. Cendyn Group and Cornish-Adebiyi v. Caesars Entertainment were dismissed for failure to plead a horizontal agreement among competitors. The key question is whether a “rim” agreement exists among the competitor users of a pricing platform, and that question is highly fact-specific. Companies should track this case law closely.

Self-reporting has become easier to navigate

The Department’s March 2026 Corporate Enforcement Policy, its first-ever policy for all criminal non-antitrust cases, carves out Sherman Act violations, leaving the Leniency Program intact. But it provides something new: a complementary path for companies to resolve non-antitrust criminal exposure (such as fraud) alongside an antitrust leniency application. For companies that may face multiple types of criminal exposure, this coordination makes self-reporting more attractive and more manageable.

Concentrated industries warrant particular attention

Housing, hospitality, healthcare, and other sectors with concentrated competitor sets and third-party revenue-management tools face acute exposure. The combination of repeat interactions among the same competitors, shared software platforms, and pooled non-public data creates conditions where a “rim” agreement may be more readily inferred by prosecutors or plaintiffs. Companies in these industries should assess whether their technology arrangements could be characterized, even if unintentionally, as facilitating coordination among competitors.

Practical next steps

Companies should take the following concrete actions:

• Inventory all pricing and procurement technology. Map every pricing, revenue-management, bid-preparation, and market-intelligence tool, both internally developed and third-party licensed. Document what data flows in, what outputs are generated, and where those outputs go. Algorithmic tools leave extensive digital trails: logs, timestamps, parameter changes, query histories. Regulators can access this data. An incomplete inventory creates blind spots that become enforcement vulnerabilities.


• Audit competitor-data inputs and outputs. Identify which tools pool non-public competitor data. Document the business justification and articulate why that pooling is, or is not, consistent with the antitrust laws. Key questions include: Does the tool receive inputs from competitors? Do outputs influence competitors’ pricing or bidding? Is there any mechanism, active or passive, for competitors to coordinate through the tool? The RealPage consent judgment targets the ingestion of real-time competitor data and the return of granular recommendations, not algorithmic pricing itself.


• Embed antitrust review in AI governance. Privacy, cybersecurity, IP, and bias programs are not substitutes for antitrust review. An AI governance program may be insufficient if no one is asking whether the tool facilitates coordination with competitors. Antitrust counsel should have authority to review data flows, vendor contracts, and use cases before AI tools are deployed in competitively sensitive areas.


• Invest in algorithmic sophistication and monitor frequently. Emerging academic research suggests simpler, less sophisticated pricing algorithms could be more prone to producing inflated pricing than more advanced designs. Algorithms that lack the sophistication to model counterfactuals or explore the competitive landscape may drift toward supra-competitive prices not through strategic intelligence, but through design limitations. Companies should invest in the people and technology necessary to ensure that their algorithm’s design reflects independent, unilateral profit-seeking objectives, and should monitor algorithmic outputs frequently to detect pricing anomalies before regulators do. The compliance imperative is twofold: 1) deploy sophisticated tools that compete rather than drift and 2) maintain ongoing oversight to catch problems early.


• Review AI provider terms of service. Most AI providers train on user inputs by default. Where competing users submit confidential economic data (e.g., pricing, capacity, supply constraints), knowing that data may inform outputs to other users, the model can become a hub. For any AI platform used in pricing or procurement, companies should ask: Does the provider train on user inputs? Are opt-outs available? Is data segregation possible?


• Train broadly. Training must reach sales, pricing, procurement, engineering, data science, and product management teams. Cover both the tools and the conversations among users, including in trade-association settings. Employees should understand that shared software provides no insulation; the question is whether they knowingly contribute confidential inputs to a system whose outputs shape competitors’ decisions. Anyone with visibility into how these systems work could be a witness or a whistleblower.


• Build rapid escalation pathways. The Whistleblower Rewards Program has created a second race between the insider and the company. A company’s path to leniency can close while general counsel is still scheduling the next meeting. Build escalation pathways that can support a leniency decision in days, not weeks.


• Protect internal reporters. Confirm that non-disclosure agreements, severance provisions, and other employment restrictions do not impede internal or governmental reporting. Train employees on the protections under the Criminal Antitrust Anti-Retaliation Act. Companies that suppress reporting or retaliate against whistleblowers face compounded enforcement risk.


• Capture communications across all channels. Ensure that ephemeral messaging apps, personal devices, and informal channels are within the compliance program’s reach, with clear preservation obligations. Digital evidence in algorithmic cases does not disappear; it multiplies.


• Evaluate third-party information exchanges. If the company exchanges competitively sensitive information through benchmarking services, industry aggregators, consultants, or market-intelligence providers, ask whether each exchange is legitimate. Document the business purpose, safeguards, and legal basis. If the answer is unclear, escalate.


• Review vendor contracts. When licensing pricing, bid-preparation, or revenue-management software, review vendor contracts for provisions that may create antitrust exposure. Understand how the vendor handles your data, data from other customers, whether competitor data is pooled, and what controls exist. Negotiate contractual protections or data-segregation requirements where appropriate.

The Division’s message is clear: Core competition principles have not changed, but the volume of digital evidence and the breadth of potential whistleblowers have grown substantially. Companies that treat algorithmic and AI-driven pricing tools as antitrust risk areas, rather than governance afterthoughts, will be better positioned for the next wave of enforcement.

For more information, please contact any of the authors.