Don't block them. Own them.

Quick Take · Lead · Put AI to Work · Looking Ahead

In this issue

Seven in ten Americans oppose having an AI data center built where they live, and nearly half are strongly opposed. That's higher than opposition to nuclear plants. The progressive answer isn't a moratorium. It's a dividend.
arXiv just banned authors who let AI write their papers without checking the output. A real accountability standard, set by an institution that didn't wait for Congress.
Tipping Point Community closed a $40M fundraise and a partnership with Anthropic to put AI tools directly into the hands of frontline nonprofits serving workers most exposed to automation.
A one-hour AI workflow for the strategic decision you've been sitting on. Use the model as a skeptical advisor, not a search engine.

Quick Take

There's an honest version of the data center fight that progressives haven't quite landed yet. The opposition is real. Seven in ten Americans oppose having one built in their area, more than oppose a nuclear plant. So "build it anyway" isn't going to be the answer. But "block them all" isn't either. AI infrastructure is going to get built. The only open question is who benefits when it does, and right now we keep arguing for the option that leaves communities with the noise and the grid load and nothing in return.

There's a better option, and even Alaska figured it out fifty years ago.

Let's get into it.

$10,000

The projected payout, per resident per year, for an 11,500-person community hosting a single 1.6 gigawatt data center under the dividend model Ben Thompson floated this week. About 3.8% of the facility's gross annual revenue, distributed to every resident every year. Same data center, different rules.

Source: Daring Fireball via Stratechery

Don't block them. Own them.

Gallup published a poll on May 13 that should be the only data point anyone in tech leadership is talking about. Seven in ten Americans oppose constructing an AI data center in their local area. Forty-eight percent are strongly opposed. That's higher than opposition to a nuclear plant in your area, which clocked in at 53%. The breakdown is the part the industry doesn't want to look at: Democrats, Independents, and Republicans all show roughly the same numbers. This is not a partisan problem. It's an industry problem.

John Gruber, who is not exactly a tech-skeptic, called it "an absolute messaging and marketing disaster for the entire tech industry." He's right, and the disaster is downstream of something specific. The communities being asked to host these facilities are being asked to absorb the grid load, the water draw, the noise, the truck traffic, and the construction footprint, and in return they get a tax break for the developer and a few years of construction jobs. That's not a deal. That's an extraction.

Ben Thompson, writing in Stratechery this week, proposed something simple. If a data center is going to draw on a community's resources to generate billions in revenue, the community should be paid for the resource. Not a one-time community benefit agreement at permit time. Not a buy-out lump sum. A recurring dividend. He ran the math on a 1.6 gigawatt facility, which is a reasonable size for the next generation of AI campuses, and got to roughly $3 billion in annual revenue. Take 3.8% of that, distribute it across an 11,500-person town, and every resident gets about $10,000 a year. Indefinitely. As long as the data center runs.

If that sounds like a fantasy, it isn't. Alaska has been doing a version of this since 1976. The Alaska Permanent Fund is a constitutionally established sovereign wealth fund built on oil and mining royalties. It currently holds about $64 billion. Every year it pays a dividend to every resident of the state. Most recently $1,000 in 2025, historically averaging around $1,600. Basic-income researchers cite it as the most concrete working example of the concept at scale. A 2024 study found it reduced the share of Alaskans living below the federal poverty line by 20 to 40%, with the strongest effect on rural Indigenous populations. And Alaska, to put it gently, is not a progressive state.

Why this matters

The progressive response to data centers has mostly been the same response we have to any extractive industry: organize the community, slow the project, demand concessions. That works, sometimes, but it produces one-off wins that have to be re-fought at every site. Issue 9 of this newsletter walked through the Brookings community benefit agreement framework as the progressive answer. CBAs are good. They are also the floor.

The dividend model is the ceiling. It treats a community's resources (land, water, grid, road, social license) the way Alaska treats oil. It says: this is a public asset, and the public should be paid for it, on an ongoing basis, every year the data center runs. A community benefit agreement gets you a new park and a workforce training program. A dividend buys every household in town a year of groceries, forever.

It also reframes the political fight. Right now the conversation is "data center or no data center," and the industry assumes it can grind that fight down site by site until it wins most of them. A dividend regime changes the question to "what's our cut?" That's a conversation communities can actually win, because once the structure exists, no incoming developer can negotiate around it.

The progressive value-add to Thompson's proposal, the part he doesn't write because he isn't a progressive, is structure. Don't run this through the developer. Run it through a public trust, modeled directly on the Alaska Permanent Fund Corporation, with constitutional or charter-level protection. Tie eligibility to real residency, not corporate domicile. Index the dividend to local cost of living. Require transparency on the data center's revenue and resource draw, so the dividend calculation is verifiable. And make community-impact metrics (water, grid load, noise, jobs, broadband) public conditions of the dividend, so the community gets the floor and the upside.

What you can do

If a data center is being proposed in your area, the public hearing is the place where the dividend frame can land. Walk in with Thompson's math and the Alaska precedent in hand. The hearing isn't only the place to oppose the project. It's the place to demand a different deal. At the state level, the question to ask your legislator is whether any bill on data center revenue-sharing or sovereign-wealth-fund structure is on the table this session. If the answer is no, that's a bill worth proposing. Indiana and Iowa have already negotiated revenue-sharing agreements with Microsoft on specific projects. The pattern is there. What's missing is a state-level legal structure that makes it the default instead of the exception.

arXiv just set a real AI accountability standard. No new federal law required.

arXiv, the academic preprint repository that's effectively the front door for new research in physics, math, and computer science, announced this month that authors who submit AI-generated papers without checking the output will get a one-year ban from the site. The trigger isn't using AI to help write. It's submitting work that contains "incontrovertible evidence that the authors did not check the results of LLM generation." Hallucinated references that point to papers that don't exist. Sentences that still contain the chatbot's own meta-commentary. The kind of slop that's been quietly poisoning the academic literature.

Thomas Dietterich, who chairs arXiv's computer science section, put it directly: "if a submission contains incontrovertible evidence that the authors did not check the results of LLM generation, this means we can't trust anything in the paper." After the one-year ban, future submissions from that author have to clear peer review at an outside venue before arXiv will take them. It's a one-strike rule with an appeal process, not a permanent ejection, but the penalty has real teeth.

Why this matters

This is what AI accountability looks like when an institution doesn't wait for Congress to act. arXiv didn't pass a law. It didn't sue anyone. It updated its own submission terms, and now every researcher in fields that publish there has to take responsibility for what they put their name on. The standard is portable. Nonprofit grantmakers could add a "verify AI-generated content" clause to their RFP requirements. State bar associations could set it as a discipline standard for legal filings. Local government procurement could require it on bid responses. None of that needs federal permission.

The federal vacuum on AI accountability is real and frustrating. It's not the only place accountability can come from.

What you can do

Look at the documents your organization receives from grant applicants, vendors, and partners. Are any of them AI-generated without disclosure? Add a single line to your next contract, grant agreement, or RFP: "By submitting this document, you certify that any AI-generated content has been verified by a human author." That sentence is enforceable. It also makes the standard normal. arXiv just did it for academic publishing. You can do it for your sector.

In brief

Google's spam policy now officially covers AI-generated answers. On May 15, Google updated its Search spam rules to make clear that AI Overviews and AI Mode are subject to the same enforcement as conventional search results. Recommendation poisoning, prompt-injection attacks, and listicles built to manipulate AI summaries can now get a site demoted or removed from search entirely. This matters because readers click through to source links roughly 58% less often when an AI Overview appears. Every manipulated AI answer is worth more now, and every enforcement action is more consequential.

YouTube opened its deepfake detection tool to everyone over 18. Previously limited to high-profile creators, politicians, and journalists, YouTube's likeness detection feature is now rolling out to all adult users on the platform. You verify your identity once with a government ID and a short selfie video, and YouTube scans uploads for unauthorized AI-generated content using your face. The tool currently covers facial likenesses only, not voice clones. Still, it's a meaningful protective layer for anyone whose face is publicly visible. Organizers, advocates, and elected officials are the most common deepfake targets, and this gives them a takedown lever they didn't have before.

Progressive AI Win

A Bay Area nonprofit just raised $40M to bring AI directly to the orgs serving AI-affected workers.

Most of the AI capacity-building money in the country is going to consulting firms, big universities, and corporate philanthropy programs that route it back through their own training pipelines. Tipping Point Community, a Bay Area antipoverty nonprofit, did something different this month. They closed a $40 million fundraise. $25 million came from Sue and John A. Sobrato and Sobrato Philanthropies, the other $15 million from their annual benefit. They paired it with a partnership with Anthropic, aimed at pushing AI tools and training directly to the frontline nonprofits already serving the workers and families who'll feel AI displacement first.

The focus is sharper than most corporate AI-for-good announcements. The program targets young people ages 15 to 24 and builds toward what Tipping Point calls "AI-safe career pathways": healthcare, infrastructure, manufacturing, energy. Sectors where demand is expected to hold through the wave of automation. The Anthropic side of the partnership is meant to put AI tools into the hands of the case managers, job-training staff, and outreach workers who are doing the actual work, "so more of their time and dollars go directly to the people who need them most."

Tipping Point CEO Sam Cobbs, in the announcement: "AI is already reshaping access to opportunity in the Bay Area. We need to act now to make that possible."

This is what civil society leading on AI looks like. Not a white paper. Not a federal task force. Forty million dollars deployed through the organizations already doing the work, aimed at the population that loses the most if we get this wrong. The Bay Area version is the prototype. The model is portable to any region with a strong community-foundation backbone and an AI lab willing to partner.

Put AI to Work

Practical ways progressives can use AI this week

Run a one-hour strategy session with AI as your skeptical advisor

We've spent past issues showing how to use AI to do things. Read PDFs, fact-check stories, build small tools, replace SaaS. This one is about using AI to think. There's a decision you've been sitting on. The thing you've drafted three different times and haven't sent. The pivot you keep almost making. The hire you can't decide whether to push for. Most of us, when we sit alone with that kind of problem, just spin. Most of us, when we ask a colleague, get back politeness or agreement. What AI is genuinely good at, and what almost nobody uses it for, is being a useful skeptic.

Here's a one-hour workflow.

Block the hour. Phone face down. One actual question to work through, written at the top of the page. Open Claude or ChatGPT. Use voice mode if you want it to feel like a conversation, text mode if you want a record.

Load the context, all of it. Tell the model what your organization does. Tell it the constraints you're working under: budget, board, calendar, political moment. Tell it what you've already tried and why those didn't work. Don't skip this step. The model can't be a useful advisor if it's working from generic assumptions.

State the decision and ask it to push back. "I'm thinking about doing X. Tell me what's wrong with this plan. Be specific. Don't be polite." If the model hedges, push: "Imagine you're the most skeptical person on my board. What do they say?"

Ask it to argue the opposite side. "Make the strongest possible case for the choice I'm rejecting." This is where it earns its keep. You will hear the version of the argument you've been brushing off, presented well, by something that has no stake in the answer.

Ask what you're missing. "What's a consideration here that I haven't named yet?" Sometimes you get nothing useful. Sometimes you get the thing you've been avoiding for two weeks.

Close by writing the next concrete action. Not the plan. The next thing on your calendar. If the conversation doesn't produce one of those, that's also useful information.

A note on tools. Claude tends to be more willing to push back when prompted to; ChatGPT tends to be more agreeable by default. If you're using ChatGPT, lean harder on the "be skeptical" framing. Either way, set the expectation up front. AI defaults to agreeing with you. The whole point of this exercise is to use it for the one thing your friends and your team are too kind to do.

From our friends

Your org deserves its own AI. Not Big Tech's.

Change Agent is a private AI platform built for nonprofits, unions, and advocacy orgs. Your data stays yours, it plugs into tools you already use (Google Drive, Slack, ActBlue), and it handles the tedious stuff so your team can focus on the mission. Starts at $35/month. Small nonprofits under $1M can apply for discounted pricing.

Learn more

Looking Ahead

A few threads to keep an eye on.

The Trump frontier-model vetting executive order Hassett floated for the May 13 to 20 window in Issue 14: as of this writing, it hasn't dropped. If it lands this week, the three-question test from last issue still applies. Blocking authority or just review? National-security-only scope, or does it cover the broader harms? Does it touch December's state-preemption order?

Connecticut Senator James Maroney's comprehensive AI bill passed the legislature this month and is now on the governor's desk. It covers frontier models, chatbots, employment, and content provenance. Lamont has signaled he'll sign. Connecticut becomes the next big test case for state-level AI regulation post-EO 14365.

And if a data center is being proposed in your area, the next planning commission or zoning hearing is the moment the dividend frame can do real work. Walk in with the Alaska comparison, the Thompson math, and the names of the Indiana and Iowa communities that already negotiated revenue-sharing deals. The question isn't whether the facility gets built. It's what your community gets in return.