"Big AI's Regulatory Capture" is a peer-reviewed paper presented at the 2026 ACM Conference on Fairness, Accountability, and Transparency (FAccT '26, Montreal), written by Abeba Birhane, Riccardo Angius, William Agnew, Harshvardhan J. Pandit, Bhaskar Mitra, Roel Dobbe, and Zeerak Talat. The lead authors are affiliated with the AI Accountability Lab at Trinity College Dublin, with co-authors at Carnegie Mellon University, TU Delft, and the University of Edinburgh. It is available as arXiv:2605.06806 (May 2026; DOI 10.48550/arXiv.2605.06806). The paper argues that regulation of the AI industry is "captured" by corporate actors across multiple dimensions and that this capture, abetted by government complicity, should be treated as an emergency.
Argument and method
The authors define their object as the capture of AI regulation, oversight, and public discourse by "Big AI" — the concentrated set of firms that, they contend, have come to exert outsized economic, political, and societal power. Using a design science research (DSR) methodology — desk research, expert exchange, literature review, and iterative design cycles — they build a taxonomy that serves as a conceptual model for mapping capture mechanisms and the narratives used to legitimize them. The taxonomy comprises 27 mechanisms across five categories.
They then construct an annotation template from the taxonomy and manually annotate 100 news articles (drawn from US and UK media queries), coding each for mechanism category, specific mechanism, supporting narrative, and excerpts. The analysis serves to both validate the taxonomy and quantify how often each mechanism and narrative appears. The study is explicitly interpretive: it does not claim to measure the total extent of capture, only to map and quantify the mechanisms visible in the sampled coverage.
Taxonomy of capture mechanisms
The five categories of the taxonomy are:
- Discourse & Epistemic Influence — shaping media, public discourse, and knowledge production through narrative framing, deception, and influence over academic and media outputs (the paper treats academic capture and media capture as sub-types). This was the single most frequently identified category.
- Elusion of law — mechanisms operating directly or indirectly against the spirit or letter of the law, including contested interpretations or violations of antitrust, privacy, copyright, and labour law.
- Direct influence on policy — lobbying, the revolving door, and related means of shaping regulation.
- Government adopting industry framing — public bodies or officials repeating or considering only industry positions while diminishing others.
- Conflation of public and private interest — presenting private commercial interests as public benefits.
Findings
The annotation identified 249 instances of capture mechanisms across the 100 articles. The distribution is concentrated: Discourse & Epistemic Influence accounted for 79 instances (32% of the total), and the ten most frequent mechanisms — half of all instances — fall within the Discourse & Epistemic Influence, Elusion of law, and Direct influence on policy categories, with Elusion of law the most frequent single category among that top set. Capture mechanisms frequently co-occurred with narratives that rationalize them. Of the 100 articles, 49 contained at least one such narrative; the most frequently invoked were "Regulation stifles innovation," "Red tape," and "National Interest."
The authors draw lessons from regulatory-capture dynamics in adjacent industries — tobacco control, fossil fuels, and the pharmaceutical/FDA context — and close with transferable tactics for uncovering, resisting, and challenging capture, including the construction of counter-narratives. They emphasize that capture is produced by "coalescing forces" of industry and government together rather than by industry acting alone, hence the paper's framing of "government complicity."
Provenance and status
The source is a 22-page academic paper distributed via arXiv under a Creative Commons license and presented at FAccT '26. As a single recent paper, the page carries medium confidence; its empirical claims (instance counts, category distribution) are specific to the 100-article sample and the authors' coding scheme, which they describe as interpretive rather than exhaustive.
Relationships
- supports: Regulatory Capture in AI Policy — anchor source for the concept's taxonomy and findings.
- related: AI Power Concentration, AI Now Institute, AI Safety vs. AI Ethics Divide