Wyden's AI Safety Net Pays Workers — and Bills Data Centres

By Daniel Rao

Let us be precise about what happened on Friday, 23 October 2026, because precision is the whole story here. Senator Ron Wyden did not pass a law, attracted no Republican co-sponsor, and did not put a price tag on anything. What the Oregon Democrat and Senate Finance ranking member did was release draft legislation — a plan, in public, for a country that is frightened of losing jobs to artificial intelligence before it has actually lost them at scale. That candour, offered up front rather than buried in a footnote, is what makes the Stability and Opportunity for Workers Act worth taking seriously. It is a safety net designed in daylight, with its gaps visible.

The draft, released alongside Senate Democratic Leader Chuck Schumer and Senators Michael Bennet and Elizabeth Warren, would rebuild American unemployment insurance from the floor up. And it would ask the industry at the centre of the anxiety — the data-centre operators whose warehouses of servers power the AI boom — to help pay for it. Democrats only, no bill number, no cost estimate, no tax rate yet. If that sounds like a thin basis for a headline, it is also the honest basis. Congress has a habit of legislating after the damage is done. Wyden is trying, unusually, to legislate before it.

"Americans feel the looming fear of AI job losses, and Congress has a responsibility to help the country prepare." Wyden said in the Senate Finance Committee release announcing the draft. It is a modest sentence, and that is its strength. It does not claim a catastrophe has arrived. It claims a fear is real, and that fear alone is a reason to prepare.

Photo: The United States Capitol in Washington, D.C. Duane Lempke, via Wikimedia Commons, CC0 1.0 Universal Public Domain Dedication. This is a file photograph of the Capitol, not a photograph of the draft legislation or its release.

What the draft would actually do

Start with the system as it stands. In most states, unemployment insurance today provides up to 26 weeks of benefits replacing roughly half of a worker's previous wages, with exact levels and eligibility varying sharply from state to state. Some states pay for far fewer weeks.

Wyden's draft would impose a national minimum standard. Every state would have to offer at least 26 weeks of benefits replacing at least 75% of a worker's average weekly wages, with state maximum benefit levels set at no less than two-thirds of the state's average weekly wage. Those figures come from the Washington Center for Equitable Growth's analysis of the draft text, not from the Finance Committee's broader summary about "new minimum standards for benefits".

The draft would also change what happens when the economy turns down. It would make 52 additional weeks of benefits automatically available in high-unemployment periods, modernising the Extended Benefits programme, which currently offers only 13 to 20 extra weeks. In a recession, in other words, the safety net would lengthen on its own, rather than waiting for Congress to pass emergency extensions in the middle of a crisis — the pattern that has defined every downturn in living memory.

Coverage would widen, too. The draft expands work-sharing, requiring every state to establish a programme; lowers earnings thresholds so part-time workers qualify more easily; adds an eligibility test extending coverage to workers misclassified as independent contractors; and expands coverage to self-employed and gig workers.

Then there is the jobseeker's allowance, the draft's answer to the people who fall through every crack. This would be a federally financed, modest weekly benefit for unemployed workers who are seeking work but would not qualify for traditional unemployment insurance — new entrants and re-entrants to the labour force with limited wage history, for example. The Finance Committee's own summary puts it plainly: the allowance exists "to ensure no one falls through the cracks, including gig workers". It is a small benefit by design. Its significance is philosophical: the system would stop pretending that only workers with a neat, continuous employment record count as workers.

The draft builds on Wyden's earlier Unemployment Insurance Modernization and Recession Readiness Act, and carries endorsements from the National Employment Law Project, Economic Security Project Action and Employ America.

The retraining bet: 26 more weeks to learn

The most interesting part of the draft is not the higher weekly payment. It is the Transition Wages programme, which answers a different question: what happens after the cheque arrives?

Under the proposal, a worker displaced by AI disruption who enrols full-time in an approved training programme would receive an additional 26 weeks of federally funded benefits — on top of the standard entitlement — to pursue education and retraining for a higher-paying job. The training would be designed to be completed within one year and to prepare the worker for an in-demand sector or occupation, including through Workforce Innovation and Opportunity Act-authorised Workforce Pell Grants.

This is the draft's central gamble. It assumes the jobs of the AI era can be trained for, that a year of full-time study is enough to reach one of them, and that the worker can afford — in time, in childcare, in rent — to spend that year studying. Those are large assumptions, and supporters should be pressed on each of them.

"The support meant to protect workers from financial devastation when they lose their jobs is brittle and out of date, and needs to be updated before the next economic crisis." Wyden said in the Finance release. Brittle is the right word. The current system was built for a world of factory layoffs and rehiring, not for a world in which an entire occupation can shrink while the economy as a whole keeps growing. A system that snaps under the second kind of shock is not a safety net. It is a decoration.

Who pays: the data-centre tax package

Alongside the unemployment bill, Wyden released draft text for a separate data-centre tax package. It would establish an excise tax on data-centre operators and remove tax breaks the industry currently enjoys, including bonus depreciation, opportunity-zone benefits and research and development tax credits.

Here the candour runs out, and the draft's authors say so themselves. The proposed tax rate remains undetermined pending revenue estimates, and no official cost estimate for the expanded benefits has been released. Wyden floated the tax idea in a white paper in August, and the October release is the draft-text version of that idea — but the numbers that would make it a funding plan are not there yet.

Photo: The United States Capitol in Washington, D.C., in August 2007. Gryffindor, via Wikimedia Commons, released into the public domain. A file photograph of the Capitol, where the prospects for any Democratic minority proposal will be decided.

Why now — and why the evidence is messier than the headlines

The timing is not mysterious. The week of the release brought a drumbeat of corporate announcements that made AI and job cuts feel like the same story. Barclays' 5,000-job cut days earlier and Amex's AI-linked cuts landed within days of each other, and both were framed, by the companies themselves, around technology-driven efficiency. We checked our own earlier coverage of both announcements before writing this piece, and the pattern they form — large, profitable employers trimming headcount while pointing at AI — is the backdrop against which Wyden's draft will be read.

But the backdrop is not the evidence, and the evidence is messier than the headlines suggest. Economists have not identified evidence of widespread, economy-wide AI-driven job displacement. A September 2026 Goldman Sachs analysis found AI-related employment pressures concentrated in a limited group of industries and workers, with younger employees potentially facing greater hiring challenges. Stanford's Digital Economy Lab, examining payroll data through June 2026, found no evidence of widespread economywide displacement — but employment among workers aged 22 to 25 in AI-exposed occupations stood 19% below where it would have been had it kept pace with less-exposed peers. That shortfall reflects reduced hiring, not increased separations, and the researchers themselves describe it as early indicators, not proof of causation.

The politics, meanwhile, are unambiguous. An Associated Press-NORC Center for Public Affairs Research poll published earlier in October 2026 found nearly 80% of Americans considered protecting workers in the AI era extremely or very important. Wyden is not inventing a constituency. He is speaking to one that already exists and is waiting, with growing impatience, for someone to do something. The context for that impatience is visible in our own reporting on how AI has topped the US layoff charts, with 120,136 job cuts cited this year — a figure drawn from employer announcements, not from proof that AI caused each cut, but a measure of how the language of layoffs has changed.

"I think it would be malpractice to not get ready, because we're in a position now where many citizens feel the looming fear of AI job losses." Wyden said, as reported by Newsmax from Associated Press coverage. Malpractice is a doctor's word, deployed by a senator, and it lands because it reframes inaction as negligence rather than prudence. It is also, to be fair, a word that invites a question: if preparation is a professional duty, what counts as preparation? A draft with no price tag is a start. It is not yet an answer.

Not the only plan in town

Wyden's draft arrives in a Congress that has been circling this problem from several directions at once, and the comparison is instructive. No single proposal has commanded a majority. Together, they map the argument.

Senator Mark Kelly's Make AI Work for Americans Act, released on 24 September 2026, takes the tax-and-spend route most directly. It would create an AI Horizon Fund, a dedicated federal trust fund financed by three taxes — on excess profits, on digital advertising services and on AI usage — to support workers, training and communities, overseen by a 15-member Artificial Intelligence Advisory Council drawn from AI experts, labour, higher education and workforce development. "Big Tech stands to make enormous profits, while workers are facing real uncertainty about what AI will mean for their jobs and their futures." Kelly said.

The transparency route has fared worse. The AI-Related Job Impacts Clarity Act, from Senators Josh Hawley and Mark Warner, would require major companies and federal agencies to report AI-related layoffs and workforce effects quarterly to the Department of Labor, with the figures publicly disclosed. Introduced last year, with a House version following in July 2026, it has stalled despite bipartisan support — a fact that tells you something sobering about this Congress. Even counting the problem, it turns out, is contested.

Senator Warner, a co-sponsor of that stalled reporting bill, captured the long-view tension in a single exchange. "I actually believe over a longer haul, five years plus, AI will probably create whole new categories of jobs." He said. "But what does that do for the interim? And we don't have good data yet." The interim is the whole ballgame. Every proposal in this debate is, at bottom, an attempt to bridge the interim — the years between the jobs that are shrinking and the jobs that have not been invented yet.

Then there is the maximalist end of the spectrum. Representative Ayanna Pressley's federal job guarantee resolution, reintroduced in September 2026 after first being proposed in 2021, calls for a federally funded, locally administered job guarantee paying at least $25 an hour, with regular cost-of-living adjustments, spanning emergency preparedness, disaster relief and recovery, childcare, senior care and public art.

Representatives Greg Casar, Valerie Foushee and Sara Jacobs took yet another route in August 2026 with the AI Tax and Work Protection Act, which would tax large AI companies on whichever is higher — the value of their tokens or their AI-product revenue — with the rate rising if unemployment rises, and use the revenue to create jobs offsetting AI layoffs. "AI CEOs are openly predicting their technology could put millions of Americans out of work, and our government is doing nothing to save jobs and protect working people." Casar said in a written statement. Casar's mechanism is the most automatic of the lot: the worse the labour market gets, the higher the tax climbs. It is also, like Wyden's, a Democratic proposal without a Republican partner.

Industry's own answer: Anthropic's 5% fund

While Congress drafts, at least one company has tried to answer the question itself. On 18–19 October 2026, Anthropic committed 5% of its shares to a new wholly owned Anthropic Public Benefit Fund, with up to an extra 2% at the board's discretion for every 50% increase in company value above a baseline. The fund could support nonprofits, direct distributions to specific populations and AI labour-market impacts — Anthropic had already committed $200 million to studying AI's effects — and its equity will not be used for lobbying or political donations. The commitment is subject to stockholder approval, and combined with founder and employee pledges, roughly 20% of the company's shares would be committed to public good.

The uphill path — and why this draft still matters

So where does this leave the Stability and Opportunity for Workers Act? On the evidence available, facing an uphill path. All of its named sponsors are Democrats, in a Republican-controlled Congress, and no Republican support for the draft has been found. The broader debate has Republicans generally favouring a hands-off approach to AI development and resisting measures they fear could discourage AI investment. Wyden himself has acknowledged, as reported from Associated Press coverage, that AI companies have not endorsed the new bill and that considerable legislative work remains. That is a polite way of saying: this is the beginning of a negotiation, not the end of one.

Here is our judgment, stated plainly. It would be colossally premature to call this draft a rescue, and shoddy journalism to pretend otherwise — a Democrats-only release with no price tag and no tax rate is a marker, not a programme. But it would be equally shoddy to dismiss it as theatre. The 75% replacement floor, the automatic 52-week downturn extension, the retraining entitlement and the gig-worker coverage are specific, costed-in-principle choices about what a modern safety net should look like, and they are now on the record for every future negotiation to reckon with. The draft's real achievement is to move the argument from whether AI will disrupt work — a question the evidence has not settled — to what the country owes workers while it finds out.

The test of this proposal will not be a vote next month. It will be whether its numbers survive contact with a revenue estimate, whether its tax package survives contact with the data-centre lobby, and whether a future Congress picks up its floor and builds on it.

Wyden AI safety net FAQ

Is Wyden's bill a law yet?

No — it is draft legislation released 23 Oct 2026, with no bill number, no cost estimate and no Republican co-sponsor.

What would workers actually get?

At least 26 weeks replacing at least 75% of average weekly wages, 52 automatic extra weeks in high-unemployment periods, plus a jobseeker's allowance for workers outside traditional unemployment insurance.

Who pays?

A separate draft data-centre tax package (excise tax + removal of bonus depreciation, opportunity-zone and R&D breaks) — but the rate is undetermined pending revenue estimates, so there is no funding total yet.

Sources: Senate Finance Committee release, 23 October 2026; Washington Center for Equitable Growth analysis of the draft; New York Times reporting via syndication summaries; Newsmax and Associated Press-derived coverage; Senator Mark Kelly's Make AI Work for Americans Act one-pager; Anthropic and Reuters reporting on the Anthropic Public Benefit Fund; TechPulse USA's earlier coverage, linked above.

Daniel Rao covers AI, labour economics and US technology policy for TechPulse USA.