What Is AI Regulatory Capture? A Plain-Language Guide
By The Agent5 founder·June 28, 2026
Regulatory capture is an old problem in a new setting: the people writing AI rules increasingly depend on the AI industry to understand what they are regulating. Here is what that means, why it matters, and how to think about what comes next.
Key takeaways
Regulatory capture happens when an industry shapes the rules meant to govern it, shifting outcomes away from the public interest and toward its own priorities.
AI is especially vulnerable to capture because of extreme information asymmetry: regulators often have no choice but to rely on the industry itself for technical expertise and safety data.
AI lobbying has surged dramatically, with more than 3,500 federal lobbyists (over one quarter of all federal lobbyists) reporting lobbying on AI issues at least once in 2025.
Capture works through many channels beyond lobbying, including the revolving door between government and industry, narrative framing, ethics washing, and funding academic research.
Resource imbalances between well-funded industry actors and under-resourced civil society groups mean the public interest is structurally underrepresented in most AI policy processes.
Imagine hiring a locksmith to design the city's security code, then asking that same locksmith to inspect every door in town. That is the core tension at the heart of AI regulatory capture, and it is not a hypothetical. It is the live political economy of one of the most consequential technologies in human history. Understanding this dynamic is the first step toward thinking clearly about who actually controls AI's future.
What Regulatory Capture Actually Means
Regulatory capture typically refers to a phenomenon that occurs when a regulatory agency created to act in the public interest instead advances the commercial or political concerns of special interest groups that dominate an industry or sector the agency is charged with regulating. The concept has deep roots in economics and political science. The concept was originally developed during the middle of the 20th century, with theories about regulatory capture developed during the 1950s, 60s, and 70s by professors of political science and economics including Samuel Huntington and Marver Bernstein. Its most famous articulation came from Nobel laureate George Stigler. In Stigler's formulation, "as a rule, regulation is acquired by the industry and is designed and operated primarily for its benefit."
Regulatory capture is a form of government failure. It happens when a government agency operates in favour of producers rather than consumers. In the AI context, those producers are labs, platforms, and the broader tech industry, and those consumers are the rest of us.
Why AI Is Especially Vulnerable
Not every industry is equally at risk of regulatory capture. AI faces a particularly acute version of the problem for several interlocking reasons.
The first is information asymmetry. Regulators often rely on the industry for detailed information about operations, which can lead to a dependence on industry insiders for knowledge and expertise. This dependency can skew regulators' perspectives in favour of the industry. With AI, this problem is amplified enormously. AI requires highly complex, technical knowledge; the general public seems to have only a limited familiarity with most of these technologies, and may be easily misled by ethics washing. Because of how machine learning itself works, detailed knowledge of the inner workings of the AI may be near-impossible to acquire.
Then there is the self-reporting problem. AI developers control both the design and disclosure of dangerous capability evaluations, creating inherent incentives to under-report alarming results or select lenient testing conditions that avoid costly deployment delays. Regulators, investors, and the public therefore face a critical information asymmetry: they must trust safety claims based on self-reported evaluations with minimal methodological transparency.
Finally, concentration matters. Regulatory capture has an economic basis: vested interests in an industry have the greatest financial stake in regulations affecting them, and so are more likely to try to influence the regulator than relatively dispersed individual consumers, each with little incentive. A handful of AI labs control most frontier development, giving them resources and motivation that no consumer group or civil society organization can easily match.
The Lobbying Surge: Numbers That Tell a Story
The scale of AI industry political activity has grown at a pace that has no recent parallel in tech policy history.
According to data compiled by OpenSecrets, 648 companies spent on AI lobbying in 2024 versus 458 in 2023, representing a 141% year-over-year increase. By 2025 the figure had grown further. More than 3,500 lobbyists, a quarter of all federal lobbyists, reported lobbying on AI issues at least once in 2025. Over the last three years, the number of AI issue lobbyists on Capitol Hill has grown by nearly 170 percent.
The dollars behind that activity are substantial. Eight of the largest tech, AI, and social media companies spent a combined $36 million on federal lobbying during the first half of 2025, an average of roughly $320,000 per day that Congress has been in session. Individual company figures show the acceleration clearly. During the first half of 2025, Meta spent a record $13.8 million on lobbying, the most the company has spent on lobbying in the first half of a year since it first hired federal lobbyists in 2009. Nvidia spent $1.6 million on lobbying during the first half of 2025, an increase of 388% from what it spent during the same period in 2024. OpenAI spent $1.2 million on lobbying during the first half of 2025, an increase of 44% from what it spent during the same period in 2024.
Why the surge? Over the last year, AI companies have started to position the success of the technology as pivotal to national security and American competitiveness, arguing that the government must therefore support the industry's growth. That reframing is itself a form of regulatory influence.
The Mechanisms: How Capture Actually Happens
Lobbying dollars are only one instrument. Researchers have documented a much wider toolkit.
A 2026 study published at the ACM Conference on Fairness, Accountability, and Transparency mapped the landscape systematically. The researchers developed a taxonomy of mechanisms enabling capture to provide a comprehensive understanding of the problem. Their taxonomy of capture consists of 27 mechanisms across five categories. Analyzing 100 news stories published around four global AI events between 2023 and 2025, they found 249 cases fitting capture patterns. One of the most prevalent was "narrative capture," which the team describes as attempts to influence the position or decisions of public officials and regulations.
The most recurring categories of mechanisms were Discourse and Epistemic Influence, concerning narrative framing, and Elusion of law, related to violations and contentious interpretations of antitrust, privacy, copyright, and labour laws. The narratives most frequently invoked to rationalize capture included "Regulation stifles innovation," "Red tape," and "National Interest."
Beyond lobbying and narrative, Big Tech is involved in the political sphere through direct consultations with the government, lobbying, hiring experts closely working with governments through the revolving door, and funding influential think tanks, universities, and experts. That last category, funding academic research, has its own name: Big Tech directly funds or employs experts and expert groups that form part of consultative interest groups around policymaking, including funding academic research for "ethics washing" and generating support, and supporting influential think tanks.
AI companies are found to evade regulatory enforcement and pressure regulators to change policy, for example, by withholding their digital services in relevant jurisdictions while at the same time establishing partnerships with government agencies to redesign digital infrastructure.
The Revolving Door in AI Policy
One of the most durable capture mechanisms across industries is the movement of people between regulatory agencies and the industries they are supposed to oversee. AI is no exception.
There is often a revolving door between regulatory bodies and the industries they regulate, where individuals move back and forth between industry jobs and regulatory positions. This creates conflicts of interest, as regulators may be more inclined to favour the industry in hopes of future employment.
In AI this plays out at multiple levels. OpenAI disclosed a new in-house lobbyist, Meghan Dorn, who worked for five years for Senator Lindsey Graham and started at OpenAI in October. Both OpenAI and Anthropic made hires over the last year to coordinate their policymaker outreach. Anthropic brought on its first in-house lobbyist, a Department of Justice alum, and OpenAI hired political veteran Chris Lehane as its new VP of policy.
The same pattern appears in energy regulation, where AI's infrastructure needs intersect with power markets. Former FERC Chairman Rich Glick's consulting firm represents the Data Center Coalition in the stakeholder process of the FERC-jurisdictional power market PJM Interconnection. Other former FERC commissioners have also gone through the revolving door to work at firms connected to the data center industry.
The Resource Imbalance: Who Can Afford a Seat at the Table
Capture does not require malice. It can emerge simply from a persistent imbalance of resources between those who want strong oversight and those who want lighter oversight.
The contrast in scale is stark. Where a major defense or dual-use firm may report lobbying in the millions of dollars per year, governance-focused organizations typically operate in the low six figures or less. This disparity in resources is reflected in the 2025 filing data, where the financial footprint of civil rights and policy-focused groups is dwarfed by the major industry players. For concrete illustration: Lockheed Martin and Microsoft reported total annual lobbying expenditures of $15.7 million and $10.1 million respectively. On the other hand, governance-focused organizations operate on a much smaller scale. The Leadership Conference on Civil and Human Rights, which lobbies on critical issues surrounding algorithmic bias and surveillance, reported $1.8 million in expenditures, and the Disability Rights Education and Defense Fund spent $50,000.
This asymmetry shapes which voices regulators hear most often, and whose technical framings they absorb.
What Is at Stake: The Public Interest Case
Regulatory capture in AI is not merely a procedural complaint. The past decade has seen a rapid growth in the development and integration of AI technologies altering virtually all societal infrastructures, such as information ecosystems, education, finance, healthcare, and law enforcement, affecting millions of people worldwide.
Pew Research found that 60% of the US population would be uncomfortable with the use of AI in their healthcare services, while 72% of respondents of a UK survey expressed a need for stronger regulation in order to feel comfortable with AI technologies. Despite the broad public support for protecting public interests through regulation of the AI industry, there are growing concerns about how the tech industry's outsized influence on policy-making may obstruct meaningful safeguards and public priorities.
The governance gap shows up concretely in how safety standards get set. Independent external scrutiny can address the trust deficit by verifying reported results, assessing whether evaluations are sufficiently rigorous to uncover real risks, and providing credible third-party perspectives on whether safety claims are justified. Yet truly independent scrutiny is difficult when the same companies defining the standards also provide the data regulators rely on.
On the legislative side, industry actors have pushed hard on preempting stronger rules. During the second quarter, Big Tech tried to cash in on years of lobbying by pushing a provision in President Donald Trump's sweeping spending bill that would have stripped states of the power to regulate AI and social media algorithms for the next 10 years. In September 2025, Meta launched a new Super PAC called American Technology Excellence Project with tens of millions of dollars earmarked to support tech-friendly candidates in state elections and oppose emerging state AI regulation.
The Agent5 Angle: Reasoning About What Happens Next
Getting smart about AI means getting smart about AI governance, and getting smart about AI governance means understanding the incentives shaping it. Regulatory capture is not a conspiracy theory; it is a documented, theorized, and historically recurrent feature of how powerful industries interact with the institutions meant to govern them.
Here is how to think probabilistically about where this goes.
One prediction: the lobbying surge is not near its peak. AI did not just increase its footprint in Washington in 2025. It ate tech lobbying whole. As AI intersects with healthcare, finance, defense, and education, the number of sectors with a financial stake in AI rules will keep growing. More sectors mean more lobbying pressure, from more directions, on more regulators.
A second prediction: state-level experimentation will continue, but face strong headwinds from federal preemption efforts backed by industry. According to the National Conference of State Legislatures, all 50 states introduced AI-related legislation in 2025, with dozens enacting measures spanning transparency, consumer protection, labor impacts, criminal law, and professional licensing. That state-level activity is also the primary target of industry preemption strategies.
A third prediction: the counter-pressure will matter more than it currently appears. Governments compete with other governments in using AI regulation, privacy, and intellectual property regimes to promote their national interests, while companies behave strategically in this competition. That multi-player dynamic creates friction that no single industry coalition can fully control. The EU AI Act, international coordination efforts, and civil society organizations that specialize in AI accountability are all forces that complicate the capture picture.
The honest probabilistic read: capture is a persistent risk, not an inevitable outcome. The degree to which it succeeds will depend heavily on whether non-industry expertise gets funded, whether regulatory bodies develop genuine internal technical capacity, and whether the public develops enough AI literacy to hold policymakers accountable for the choices they make. That last part is exactly what knowing how to get smart about AI is for.
Not in most cases. Lobbying, funding think tanks, participating in regulatory consultations, and hiring former government officials are all legal activities in most jurisdictions. Regulatory capture describes an outcome, not a crime. It becomes a legal matter only when specific rules around corruption, disclosure, or conflict-of-interest are violated.
Ordinary lobbying is an industry trying to influence policy on a specific issue. Regulatory capture is a deeper structural condition in which the regulated industry shapes not just one law but the ongoing institutional capacity and framing of the regulator itself, including what expertise regulators rely on, what narratives frame the debate, and what standards count as evidence of safety.
Ethics washing refers to a company or industry group publishing high-profile voluntary ethical principles, responsible AI frameworks, or safety commitments in ways that create the appearance of meaningful self-governance without accepting binding obligations or independent enforcement. Researchers have noted that funding academic ethics research can serve a similar function, lending credibility to industry positions while limiting independent criticism.
No. Regulation can and often does serve the public interest effectively. Regulatory capture describes what happens when the process goes wrong, not the inevitable result of any regulation. The risk is that rules end up serving incumbents over the public: for example, by raising compliance costs that small competitors cannot afford, or by locking in voluntary safety standards that the largest companies already meet.
Yes. The EU AI Act, which entered into force in 2024, establishes a legally binding risk-based framework with significant penalties for non-compliance, which creates a harder regulatory floor than the largely voluntary US approach. However, researchers have documented that industry lobbying shaped the EU AI Act as well, moderating some of its original requirements. No regulatory system is immune; the degree of exposure differs.
Awareness is the first step. Public comment periods on AI regulation, support for civil society organizations working on AI accountability, and demand for mandatory disclosure of industry funding in regulatory advisory processes are all levers that exist within democratic systems. Developing personal AI literacy also matters: the more the public understands what these systems actually do, the harder it becomes for any actor to dominate the debate through technical mystification.
