
The fight over AI policy is converging on a single demand: accountability. The public wants a share of the gains and a say in how the technology is built. Policymakers, especially in the states, are pushing for someone to answer for the harms. Washington is unlikely to agree on one rule to deliver either. The rules that do arrive will test whether organizations can explain their data and prove what their AI did and why.
That is the picture Politico senior tech reporter Brendan Bordelon painted in a recent episode of AI Radicals, recorded the week after the midterms. Host Satyen Sangani asked him why data centers became the place voters took their AI anxieties, where Democrats might take regulation, and what any of it means for the people building and deploying the technology. Bordelon says he has covered tech for about 10 years, and his forecast comes with plenty of caveats. For data and analytics leaders, the direction of travel is still clear enough to plan around.
The AI backlash is a demand for a share of the gains
Start with why the pressure exists. Bordelon argues that this was the first election where the data center issue was clearly a driving factor, though he acknowledges there is no clear data on how much it moved voters. His strongest evidence is on the ballot. By his recollection, 41 or 42 data center measures appeared on ballots across states and localities, and he is fairly confident that all but one went against development. The record largely bears him out. On November 3, voters in 10 states decided 41 local data center measures, 28 of which would have limited development and 13 of which would have facilitated it.¹ By mid-November, Ballotpedia’s tally for the year showed 46 of 47 decided measures limiting development and one favoring it.² He ties the result to a broader sense among voters that decisions about AI are being made by a small circle of executives and officials without their input.
Sangani identified what sits underneath. Headline indicators for inflation and unemployment may look healthy, yet the gains from AI and asset markets have accrued to very few people, while many things people once afforded are now priced out, from concierge medicine to private school tuition to event tickets. Bordelon agreed and added that the frustration extends well beyond AI. Data centers absorb the backlash because they are physical and vulnerable.
The policy conversation is already adapting. Bordelon describes growing interest in a ratepayer return, an idea that goes beyond protecting households from higher electricity bills by paying communities that host a data center for the externalities they absorb. The logic is simple: if companies capture the upside and everyone else carries the costs, the people carrying them will ask for a cut.
Enterprise leaders should recognize the pattern. Inside organizations, employees, customers and business units ask the same question about AI programs: who benefits, and who bears the risk? Evidence of the gap is easy to find. MIT’s Project NANDA, in its widely cited GenAI Divide report, found that only about 5% of integrated AI pilots were extracting millions in value, while most remained stuck with no measurable profit-and-loss impact.³ It was a preliminary study based on interviews, surveys and a review of public initiatives, and its authors flag sample limitations, so read it as a signal rather than a universal failure rate. Programs that cannot show where the value went will find support eroding long before regulators arrive.
Washington will probably stall, and the states will not
If public pressure is building, the natural assumption is that Congress will respond. Bordelon is skeptical. He describes the incoming Democratic majorities as large, but says the party has no unified approach to AI regulation because it has not had to write AI policy before. He sketches three camps: a relatively industry-adjacent group favoring third-party auditing and emergency government powers, a safety-focused wing calling for bans on superintelligence or recursive self-improvement, and a group concentrating on nearer-term harms such as bias in hiring, credit and housing. In his telling, a presidential veto, a Senate that still needs 60 votes, and a calendar crowded with tariffs, the Iran war and investigations all work against action, and his conclusion is blunt: the likeliest federal outcome is still probably nothing.
That forecast does not mean a regulatory vacuum. Bordelon says many advocates are no longer putting all their eggs in the federal basket and are turning to the states. California’s SB 53 already requires frontier AI developers to publish safety frameworks, report certain critical safety incidents to the state, and protect whistleblowers.⁴ Illinois and New York have since followed with frontier AI safety laws of their own, and Illinois goes furthest, requiring large frontier developers to retain an independent third party to audit their compliance every year beginning in 2028.⁵ Bordelon expects larger states to join in, helped, in his telling, by newly elected Democratic governors and legislatures. Texas, for one, elected Democrat Gina Hinojosa as governor on a platform that included a data center moratorium.⁶ He suggests that a critical mass of state laws might eventually push Congress and the White House to act.
Sangani raised the obvious complications. Open source models blur questions of responsibility, and any rule that slows US labs invites competition from China. Bordelon does not dispute the tension and notes that many safety advocates share the worry about China. He also believes some questions, particularly those touching national security, have to stay at the federal level. In the meantime, the realistic planning assumption is a patchwork. Organizations that operate across states will face overlapping requirements that change over time, and none of them can be reliably predicted.
The next rules will ask who is liable, and an auditor may be inside the building
The content of those state rules matters as much as their number. Bordelon expects the next wave to move toward independent verification organizations: outside groups with AI safety expertise that, as he describes them, would be embedded inside labs with employee-like access and report to the state if they find problems. California has taken the first step. On September 9, Governor Newsom signed SB 813, which creates a framework for state-credentialed verification organizations, and AB 1405, which creates a registry of AI auditors with independence standards.⁴ Neither law requires developers to obtain an audit,⁷ and onsite embedding is still a proposal: a follow-up executive order asked the state to develop recommendations on requiring it.⁸ Bordelon notes that most people inside the labs welcome the idea, since they want someone checking the work.
The second thread is liability. When a model causes damage through a cyberattack, financial loss or harm to people, who answers for it, and in what capacity? Bordelon explained why he thinks states will engage:
“If the attack takes place in the state or if people are harmed from that state, I think that the sense is that there is a jurisdiction there. It’s not as clean as if it was a federal liability standard, but I do think states are gonna start talking about that as well,” Bordelon explains.
He adds that some measures, such as banning model releases or recursive self-improvement, are national security questions that have to stay at the federal level, and that the labs themselves believe only DC can handle their biggest worries. Of the labs’ preference, he says: “Ideally, they would want to see a federal approach, but in the interim, there’s a lot that the states can do and there’s a lot that they’re already doing.”
Read those developments as an enterprise data leader would. Embedded auditors, whistleblower channels and liability regimes all converge on the same underlying questions. What data went into this model? Where did it come from? Who approved its use? How did it reach this output? Call this provenance. A company that can answer quickly has a defensible position. A company that has to reconstruct the answer from scattered documentation and institutional memory does not.
What a regulation-proof data foundation looks like
No one can know which rules will survive, and that uncertainty argues for a foundation that works under any of them. Four capabilities do most of the work.
The first is visibility. A data catalog that documents what your organization has, who owns it and how it is used gives auditors and internal stakeholders a concrete place to start. If you are building that discipline from scratch, our primer on what a data catalog is covers the fundamentals.
The second is traceability. Data lineage shows how data moves from source to model to output, which turns “what did this model consume?” from a weeks-long investigation into a lookup. When a regulator, an embedded auditor or an internal risk team asks about a specific result, lineage supplies the evidence trail.
The third is accountability. Strong data governance assigns ownership, defines policy and records approvals, and it extends naturally into AI governance as models and agents join the picture. Teams that treat what data governance is as an operating discipline, rather than an annual compliance exercise, adapt faster when requirements change. Identifying and managing your critical data elements gives that discipline a practical starting point, and attention to data privacy matters as bias and discrimination concerns, which Bordelon flags in hiring, credit and housing, move into law.
The fourth is trusted delivery. Governed data products give business users and AI systems vetted inputs, with ownership and quality signals attached. That addresses the shared-gains question from the start of this piece. When teams can see which data products power which AI use cases, they can show where the value was created and who contributed to it. The same principle underpins trusted data for AI agents, where autonomy only works if the underlying data deserves trust. Organizations building toward this often benefit from a clear data product operating model.
The business case is straightforward. Investing in these capabilities lowers risk by shortening audit and response time, speeds delivery by reducing rework when requirements shift, and protects the credibility of AI programs with the stakeholders whose support they need.
Questions to ask before the first rule lands
You do not need to predict the legislation to start preparing. Three questions give a practical read on where you stand. First, can you trace any model output back to its data sources and the people who approved them? Second, who in your organization would answer a regulator or an embedded auditor on short notice, and do they have the tools to do it? Third, could you show business stakeholders, in plain terms, where the value from your AI investments went and who contributed to it?
If any answer is shaky, that is the work to prioritize. The rules are still being written, and the organizations that prepared early will have fewer surprises when the details arrive.
The direction is clearer than the destination
Bordelon is careful to say that outcomes in Washington remain uncertain, and his candor is useful. What the conversation with Sangani makes plain is that demand for accountability is growing from voters, states, labs and enterprise stakeholders at once, even as agreement on the specific rules remains out of reach. Companies cannot control that process. They can control whether they are ready to explain their data when someone asks.
Curious to learn how a governed data foundation can support your AI strategy? Book a demo with us today.
Sources & notes
Every external claim on this page is independently verifiable. The public sources are listed here.
41 local data center ballot measures in 10 states on Nov. 3, 2026; 28 would limit development, 13 would facilitate it. Ballotpedia News ↗ https://news.ballotpedia.org/2026/11/03/voters-in-ten-states-will-vote-on-data-center-related-ballot-measures-november-3/
As of Nov. 13, 2026, 47 data center measures decided in 2026; 46 limited development, one favored it. Ballotpedia News ↗ https://news.ballotpedia.org/2026/11/15/voters-have-decided-on-47-measures-related-to-data-centers-in-2026-in-december-they-will-consider-four-more/
“Just 5% of integrated AI pilots are extracting millions in value, while the vast majority remain stuck with no measurable P&L impact.” MIT NANDA, The GenAI Divide: State of AI in Business 2025, July 2025 (preliminary findings; PDF copy hosted by Webber Wentzel, provisional until MIT-hosted copy is linked) ↗ https://webberwentzel.com/News/Documents/2025/MIT1757411281972.pdf
SB 53 requires frontier AI developers to publicly disclose safety frameworks, report specified critical safety incidents to the state and protect whistleblowers; SB 813 and AB 1405 signed Sept. 9, 2026. Office of Governor Gavin Newsom ↗ https://www.gov.ca.gov/2026/09/09/governor-newsom-signs-first-in-the-nation-ai-safeguards-to-protect-californians-calls-on-the-federal-government-to-do-its-part/
Illinois Artificial Intelligence Safety Measures Act signed July 6, 2026; annual independent audits for large frontier developers from Jan. 1, 2028; Illinois is the third state after California and New York to impose transparency, safety and reporting obligations on large AI developers. Skadden, July 22, 2026 (law-firm alert; statute text not opened, provisional) ↗ https://www.skadden.com/insights/publications/2026/07/illinois-enacts-ai-safety-law-becoming-first-state
Texas governor-elect Gina Hinojosa (D) campaigned on an immediate data center moratorium. Renewable Energy World, Nov. 6, 2026 ↗ https://www.renewableenergyworld.com/energy-business/policy-and-regulation/voters-overwhelmingly-rejected-data-centers-in-the-midterms-heres-where-developers-will-have-a-tough-time/
SB 813 and AB 1405 do not require developers to obtain an audit; SB 813 creates a voluntary tier of credentialed IVOs. Crowell & Moring via Mondaq, Oct. 16, 2026 (law-firm summary; bill text not opened, provisional) ↗ https://www.mondaq.com/unitedstates/new-technology/1855868/california-establishes-first-in-the-nation-framework-for-verifying-independent-ai-auditors
Executive Order N-9-26 directs the Government Operations Agency to convene experts to develop recommendations, including on requiring onsite embedded IVOs. Office of Governor Gavin Newsom, Sept. 18, 2026 ↗ https://www.gov.ca.gov/2026/09/18/governor-newsom-issues-executive-order-to-accelerate-independent-oversight-and-advance-the-creation-of-an-ai-kill-switch/
- AI
- Alation News
- Data Governance
- Digital Transformation
Keep reading
More from the data desk



