- Lead. A peer-reviewed study by researchers at four universities — Edinburgh, Trinity College Dublin, TU Delft, and Carnegie Mellon — catalogued 249 documented cases of AI companies using the same tactics that Big Tobacco, Big Pharma, and Big Oil deployed to delay and dilute regulation of their industries.
- Fact. Analyzing 100 news articles covering four major AI policy events between 2023 and 2025, the researchers identified 27 distinct patterns of “corporate capture” — a process by which regulators come to act in the interest of corporations rather than the public.
- Stake. The findings arrive as the EU pursues antitrust action against major AI companies, the US Congress debates AI governance, and the AI industry’s lobbying spend has reached levels that dwarf earlier periods of tech sector regulatory engagement.
The paper, titled “Big AI’s Regulatory Capture: Mapping Industry Interference and Government Complicity” and published with DOI 10.48550/arXiv.2605.06806, was reported publicly by The Register on May 18. Its lead authors include Dr. Abeba Birhane, Director of Trinity College Dublin’s AI Accountability Lab, and Dr. Zeerak Talat, a Chancellor’s Fellow at the University of Edinburgh School of Informatics. The work will be presented at the ACM Conference on Fairness, Accountability, and Transparency in June 2026.
The two dominant tactics
Of the 27 identified patterns, narrative capture and elusion of law were the most frequently documented. Narrative capture involves framing regulation as innovation-stifling “red tape” — a construction that positions the companies’ commercial interests as synonymous with technological progress and public benefit. The researchers note this framing has been adopted not only by AI companies themselves but by government officials who have moved through the revolving door between industry and public service.
Elusion of law — the second most common pattern — covers “violations and contentious interpretations of antitrust, privacy, copyright and labour laws,” the paper states. This category encompasses the use of jurisdictional ambiguity, forum shopping, and deliberate statutory overreach in training data collection, intellectual property use, and employment classification. The researchers cite 249 total instances across the four events they analyzed, meaning an average of more than two documented cases per article examined.
Revolving doors and political donations
Beyond the top two patterns, the study catalogues a standard toolkit: lobbying campaigns targeting specific legislative provisions; retaliation against whistleblowers, researchers, and lawmakers who advocate for stricter rules; equity ownership arrangements that give regulated companies financial interests in the regulatory bodies nominally overseeing them; and political donations structured to build influence across multiple legislative sessions rather than around a single vote.
The revolving-door dynamic — in which former government policymakers move into advisory or employment roles at leading AI firms — is documented as both a capture mechanism and a deterrent to enforcement. Officials who anticipate post-government careers in the industry have reduced incentive to impose binding constraints on their prospective future employers, a pattern the researchers describe as structurally identical to what occurred in tobacco, pharmaceutical, and oil regulatory environments.
The regulatory context
The study’s publication timing coincides with an active enforcement period. The EU’s antitrust investigation into Big Tech’s AI operations — examining how companies may entrench market power through control of training data, compute, and model distribution — is proceeding alongside the delayed implementation of the AI Act’s high-risk provisions. In the United States, federal antitrust enforcers and state attorneys general have developed conflicting approaches, creating the ambiguity that elusion-of-law strategies are designed to exploit.
The researchers recommend separation of public and private interests through binding conflict-of-interest rules, modelled on the oversight frameworks eventually imposed on tobacco and pharmaceutical companies after decades of industry interference. The comparison carries a tacit argument: those frameworks arrived late, after significant harm had accumulated, and the question for AI governance is whether policymakers are willing to move faster than their predecessors.