OpenAI is spending political capital on a race two years away. In a sweeping essay on its Global Affairs blog, the company’s chief policy officer Chris Lehane declared 2028 “the AI election” and outlined a policy playbook aimed at voters who increasingly blame artificial intelligence for job anxiety, data-center sprawl and a tech consensus that has curdled.
The pitch borrows openly from Franklin Roosevelt’s New Deal. Researchers who study AI policy have answered with accusations of historical amnesia and corporate self-interest. This article unpacks what the company is actually proposing, where its own record undercuts the rhetoric, and which signals to watch between now and November 2028.
Table of contents
OpenAI Is Betting On The 2028 Election
The essay, published on the Global Affairs blog, reads less like a product note and more like a campaign briefing. Lehane frames the next two years as a contest over who can credibly promise that AI will work for ordinary people, and he positions OpenAI as a willing co-author of that promise. The company wants voters to associate its name with worker security rather than job displacement.
The timing is not accidental. Democrats just flipped seats in Congress by running on cost of living, and candidates who flirted with defending AI infrastructure found themselves on the wrong side of local opinion. Data-center opposition turned out to be bipartisan, and polling shows enthusiasm for artificial intelligence lagging far behind the capital flowing into it.
For the company, that political weather is a business risk. Its buildout depends on billions in financing, energy agreements and permits that elected officials can slow down or speed up. The company has already learned that branding can draw fire: its AI compute campuses rebrand was an attempt to soften exactly the kind of backlash this essay now addresses head-on.
There is a simpler reading available too. The company faces a regulatory environment where a single governor, a single state attorney general, or a single congressional committee can complicate billions in commitments. Writing worker-friendly prose is cheap insurance against expensive hearings, and framing the company as a stakeholder in political stability is itself a bid for protection. When a company starts drafting campaign issues, it is usually signaling that it wants something specific from the people who will hold the pen.
So the wager is that the company can move from villain to co-builder in two election cycles, and that voters will credit the promise before they audit the record. Whether the policy detail matches the rhetoric is the question the rest of this piece examines.
A New Deal Analogy With Missing Pieces
Lehane invokes Franklin Roosevelt deliberately. The New Deal redefined the government’s role after economic collapse, and Lehane suggests AI creates a similar inflection point: a general-purpose technology that will redistribute work and wealth unless policy shapes the outcome. He calls for a “compelling third path” between alarmism and laissez-faire.
Critics note that the original New Deal responded to a crisis by constraining powerful interests: banking rules, antitrust vigor, and legal protection for unions. The company’s version, by contrast, is authored by one of the firms at the center of the disruption, and it asks government to subsidize adoption rather than discipline concentration.
Kate Brennan of the AI Now Institute told TechCrunch the framing was “ahistorical at best and harmful at worst,” arguing the entire plan is predicated on mass AI adoption happening on schedule. Her preferred levers are blunter: revise corporate taxes, incentivize human hiring directly, and curb market power in the AI market itself.
It is also worth stating plainly what both plans share: a belief that government should be an active participant in technological change rather than a spectator. The disagreement is over who sits across the table when the terms get set. The 1930s settlement emerged from a broad, messy coalition of unions, farmers and reformers. Lehane’s invitation, as drafted, has one chair clearly reserved for the company writing it, and a few folding chairs for everyone else.
The comparison remains useful precisely because the contrasts are so visible. Here is the side-by-side between the pitch and history that critics keep pointing to:
| Dimension | New Deal (1933) | Lehane Plan (2026) |
|---|---|---|
| Trigger | Depression after a decade of excess | Anticipated disruption before mass adoption |
| Who writes the rules | Government and reform coalitions | A leading AI company drafts and promotes |
| Primary lever | Regulation plus public jobs | Adoption incentives plus tax tweaks |
| Target of discipline | Finance and monopolies | Almost none; incumbencies largely spared |
| Time horizon | Emergency legislation in years | An election-cycle narrative in 2028 |
The Five Priorities In Lehane's Playbook
The essay organizes the company’s agenda into five priorities: worker protections, education, AI safety, data-center development, and tax-and-ownership ideas. Together they form a surprisingly specific list, and each one maps onto a live legislative fight in Washington or the states.
On workers, the essay floats tax incentives for teaching AI skills, easier capital access for firms that invest in their people, and even employee ownership or profit-sharing experiments. None of it requires the kind of corporate restructuring that more radical proposals demand, which is precisely the design: keep the firms intact, nudge their incentives.
On education, the pitch leans on access. Lehane argues that AI tutors can close gaps for “children in rural communities” who lack the specialists that wealthy districts buy outright. Critics reply that access arguments ring hollow when adoption is uneven and device budgets are being set by the same companies selling the tools.
Safety is listed as a pillar, which is where the essay invites the hardest cross-examination, because OpenAI’s voting record in Sacramento does not match its current language. The data-center pillar is the most explosive, and it gets its own section below. Below is how the playbook’s attention distributes across the five pillars in this article’s reading of the essay.
The list also skips questions the skeptical wings of both parties keep asking. Nothing in the five pillars addresses workers already displaced, only those not yet arrived. There is no mechanism for community consent before a campus breaks ground, and no discussion of who pays when an adoption bet fails. A playbook this short is necessarily a set of opening bids, not a contract.
The takeaway is that the list is wide but shallow on enforcement: four of five pillars rely on incentives that the company itself could benefit from.
Illustrative weighting: how much room each pillar gets in the essay
Data Centers Are The Live Wire
If one issue will decide the AI election, it is the ground war over data centers. Towns from Virginia to Iowa have organized against energy-hungry campuses, and state legislatures have floated moratoriums. What began as a left-wing complaint has become a right-wing talking point wherever utility rates climb.
Lehane’s essay treats this opposition as a civil liberties question. He compares supporting moratoria to the rush toward the 2002 Iraq vote: a decision made hastily, on incomplete information, with consequences nobody fully priced. It is an audacious move, likening permitting protests to war fever, and few lines in the essay drew more immediate pushback.
The industry’s own numbers make the fight harder to defuse. Data-center investment announcements now rival infrastructure bills, yet the jobs figures attached to them are tiny by comparison: a few hundred permanent positions for campuses worth billions. That ratio, enormous land and power claims against modest payrolls, is the single most quotable line in every local campaign against hyperscale construction, and no policy essay has fixed it yet.
The politics here are exactly what the industry fears. Our coverage of the 2026 midterm map showed candidates losing districts for defending the rollout, and New York’s data-center debate demonstrated that even progressive politicians now treat energy and land claims as bread-and-butter issues.
Energy is the accelerant of the whole fight. Campuses pull hundreds of megawatts, utilities spread the grid upgrades across ratepayers, and the compute companies keep the revenue, which is precisely the transfer that organizers keep putting on local ballot measures. Every OpenAI campus therefore needs three friendly institutions at once: a county commission, a utility commission, and a governor willing to be photographed cutting the ribbon.
OpenAI’s ask is procedural: faster permitting, clearer energy rules, and a national frame that counts compute as infrastructure. But infrastructure only wins when nobody’s electricity bill becomes the campaign ad, and the industry has done little to prevent exactly that ad.
Classrooms, Companions And Child Safety
Education is where OpenAI’s adoption argument is easiest to sell and hardest to police. Districts are simultaneously courting AI tutors and drowning in concerns about attention, dependency and data. Lehane leans toward the courts of public opinion, promising age protections and parental controls while warning that opting out condemns rural students to a two-tier system.
The classroom fight is already reordering the industry. Google learned how sharp the backlash cuts when it offered schools an opt-out from AI that still cost them Search, an offer that read like coercion and handed challengers a recruiting speech. Apple went the other way, restricting AI companions to users 18 and up after weeks of scrutiny over child safety.
The debate also has a procurement dimension. Districts increasingly buy AI through device-management contracts negotiated at the state level, which means the companies that write the model procurement language get to define what “safe” and “effective” mean before any local school board votes on it.
The company’s own record complicates the promise. Its products court younger users, and its research arm publishes warnings about emotional dependency even as its consumer app encourages daily chat. Age verification is now on the company’s policy menu, but the engineering reality behind it remains thin.
The honest summary: the firm is arguing that child safety and AI access are complements, while parents increasingly experience them as trade-offs. Which framing wins will shape district procurement long before any 2028 referendum.
OpenAI Safety Promises Meet A Tough Record
Safety is the pillar with the longest paper trail. In 2023, OpenAI co-signed an opposition letter against California’s SB-1047, the flagship attempt to impose safety obligations on frontier developers. The company spent political capital killing a bill whose stated goal, preventing catastrophic misuse, it now claims to champion.
The record since then is more mixed than the essay admits. The company has publicly backed strengthening California’s successor safety law, and it joined Anthropic in supporting embedded safety evaluators inside labs, a proposal that raises its own independence questions.
Yet the fundamentals remain unpublished. Frontier labs, the company included, still will not say how they would contain a rogue model. The kill-switch language in campaign essays is easier to print than the containment plans it refers to.
Brennan calls the posture shift “disingenuous,” and the table below summarizes why skeptics reach for that word:
| Moment | Public stance | Critics’ read |
|---|---|---|
| 2023, SB-1047 | Co-signed opposition | Rejected binding safety law |
| 2024-25, state bills | Lobbied for federal preemption | Sought weaker rules, faster |
| Mid-2026, successor bill | Backed strengthening it | Rebranding toward the middle |
| Sept 2026, embedded evaluators | Open to hosting them | Independence unresolved |
| Nov 2026, “2028” essay | Safety as campaign pillar | Containment plans still unpublished |
OpenAI's Critics Are Not Convinced
The sharpest objection to OpenAI’s playbook is not moral but analytical. Brennan’s core claim is that the playbook’s benefits are contingent on an adoption wave that has not arrived, so every promised worker dividend is hypothetical while every conceded subsidy, permit or liability shield is real the day it passes.
Second, critics argue the plan treats market concentration as a feature. OpenAI drafts policy as a market leader negotiating the rules of a field it intends to own. A worker-first agenda that leaves procurement power concentrated looks, to skeptics, like a labor law written by management.
Third, there is the trust problem. The same organization that fought California in 2023 now lectures California on urgency. When the essay asks voters to believe in a third path, they are being asked to weigh future promises against a recent, documented record, and to grade the promises generously.
What the critics want instead is concrete: corporate tax revision, direct incentives for hiring humans, and antitrust attention to the compute market. Notably, none of those require trusting a single company’s intentions, which may be the exact reason the playbook avoids them.
There is also a practical point: policy written in a California boardroom tends to age badly. Candidates will arrive in 2028 with their own AI agendas, and a playbook designed to court both parties may end up belonging to neither. The company is trying to set an agenda that politicians adopt on their own terms, which means the company absorbs the blame for whichever parts of it poll worst while claiming credit for the rest.
What To Watch Before 2028
Five tells will reveal whether OpenAI’s bet is working. First, state legislation: does the company publicly support at least one binding safety or data-center bill with teeth before the primaries? Its stance on state data-center bills is the cleanest scoreboard.
Second, money: watch whether OpenAI’s political spending follows its prose, because policy essays without donor activity are message tests, not commitments. Third, adoption data: if mass adoption stalls, the playbook’s promises become unfundable and the story flips.
Fourth, the classroom: school-district deals and age-verification rollouts will show whether the child-safety pillar is engineering or press release. And fifth, containment: if the kill-switch talk becomes a published plan with third-party review, the critics’ central charge weakens.
The industry’s other option is the one it has chosen before, voluntary safeguards trumping public concerns, a playbook that has repeatedly collapsed under scrutiny. The 2028 election will be a referendum on whether that cycle finally broke.
The stakes extend past one company. Whoever defines the AI election frame first gets to set the overton window for 2028: whether the debate is about owning the gains of automation or merely regulating its worst excesses. Third-party test scores, state-level laws, and turnout in energy-exposed districts will all feed that definition.
One more signal: watch the language itself. If the company’s essays start sounding less like opposition research against its critics and more like a governing platform that other institutions repeat, the rebrand is working. If the company stops publishing them after the first primaries, the experiment failed, and 2028 will have shaped the AI industry far more than the industry shaped 2028.
OpenAI did not respond to requests for comment beyond its essay. The essay itself did not disappear. Both facts are data points for the story still being written.
References
TechCrunch: OpenAI is already trying to shape the 2028 election
OpenAI Global Affairs: Special Edition – 2028: The AI Election
OpenAI: Proposals for the US AI Action Plan
TechCrunch: Everyone can find a reason to dislike data-center construction
TechCrunch: Apple restricts dating apps and AI companions to users 18 and up
TechCrunch: Google will let schools disable its AI tools, but not without sacrificing Search
TechCrunch: OpenAI says California should strengthen its AI safety bill
TechCrunch: Frontier AI labs still won’t say how they’d contain a rogue AI model
TechCrunch: OpenAI’s opposition to California’s AI law makes no sense, says state senator
TechCrunch: Anthropic and OpenAI want to embed safety evaluators – will they really be independent?
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
