Who gets the leverage?

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

In this issue

A major MIT study just ranked AI's risks and asked the question that matters most: who is exposed, and who is responsible for fixing it. The answer hands progressives a mandate, not a reason to panic.
The biggest federal AI bill yet would freeze state AI protections for three years. That is the fight worth watching, and the side worth being on.
A Michigan Senate candidate is betting that workers want a real AI jobs plan, not just fear.
The most hopeful AI story of the week is happening in federal courthouses, where people who could never afford a lawyer are getting into court for the first time.
A Put AI to Work you can run this week: turn that MIT risk study into a one-page risk profile for your own community.

Quick Take

Two stories this week look like opposites and are actually the same story. A big MIT study mapped everything that could go wrong with AI. And in courthouses around the country, people who could never afford a lawyer are filing their own cases with AI and getting heard. Risk and benefit, landing on the exact same people: the ones with the least power in the room.

That is the whole ballgame. AI is not going to sort itself into fair outcomes. The job is to make sure the people most affected get a say, and a share, and some leverage. Let's get into it.

74%

The share of low-income households in America that hit at least one civil legal problem in a year: an eviction, a benefits denial, a custody fight, a wage that never showed up. Most of them face it alone. The Legal Services Corporation, which tracks this, found that low-income Americans get no help or not enough help on the large majority of these problems. For decades that gap was treated as a fixed fact of life. This week there is real evidence that AI is starting to close it, which is the most hopeful thing in this issue. More on that below.

AI's risks land hardest on the people with the least power. So does the fix.

MIT's FutureTech group, working with researchers at the University of Queensland, just published one of the most serious attempts yet to map what AI could actually do to people. They put 272 AI experts from 37 countries through a structured, three-round process and had them rank 24 distinct AI risks: not the sci-fi ones, the real ones, across jobs, finance, information, security, and more. The sobering headline: without real mitigation, the experts judged 18 of those 24 risks to be more than 10% likely to cause catastrophic harm.

But the part worth sitting with is not the ranking. It is the two questions the study asked alongside it: who is most exposed to each risk, and who is most responsible for addressing it. That second question is the one the industry usually skips. The study put it on the table and pointed where the responsibility actually sits: with the people building and deploying these systems, not the people on the receiving end.

Why this matters

Here is the quiet pattern in that data. The people most exposed to AI's harms are, over and over, the people with the least power to do anything about them: workers in automated industries, low-income users of AI-run systems, communities that never got a vote on whether these tools got pointed at them. Exposure is high exactly where leverage is low.

That sounds bleak. It is actually a mandate, and it is one progressives are built for. The gap between who is exposed and who decides is the oldest problem in organizing, and we have a century of tools for closing it: worker voice, community oversight, public accountability, the simple insistence that the people affected by a decision get a seat at the table where it gets made. The MIT study, without meaning to, just wrote the case for all of it. The answer to "the public is exposed" was never "build less AI." It is "give the exposed (much) more power."

What you can do

Pick one place AI already touches your community and ask the two MIT questions out loud: who here is exposed, and who is responsible? A school using AI to flag students. A county screening benefits applications with an algorithm. An employer running automated scheduling. You do not need a policy shop to start. You need to name who is on the receiving end and who owns the system, because almost nobody is asking, and the asking is where accountability starts. The Put AI to Work below turns this into a one-afternoon exercise.

The fight to watch: a federal AI bill that would freeze state protections

The biggest piece of federal AI legislation yet arrived this month: the Great American AI Act, a 269-page bipartisan discussion draft from Representatives Jay Obernolte and Lori Trahan. Some of it is genuinely substantive, including a public framework requirement for the largest AI developers and a federal standards center. But the provision doing all the work is a three-year preemption of state laws regulating AI development. In plain terms: for three years, states could not pass their own rules on how powerful AI gets built.

That is why labor unions, consumer advocates, and even a House Democratic commission came out against the draft almost immediately. The state level is where nearly every real AI protection of the last two years actually happened, in places like Illinois, Colorado, and California. Freezing that is not a neutral "let's wait for federal rules" move. It clears the field at the exact moment communities were starting to win.

What you can do

If you are in a state that has passed or is considering AI protections, tell your members of Congress that preemption is the line. A federal floor is welcome. A federal ceiling that erases state wins is not. State legislators and advocates should be on record now, while this is still a discussion draft and the language can still change.

A candidate treating AI as a jobs plan, not a scare

Michigan Senate candidate Mallory McMorrow is making a bet most politicians are still too nervous to make: that workers want an actual plan for AI, not just a warning about it. Her platform includes a federal apprenticeship program to train early-career workers for jobs that hold up against automation, plus expanded unemployment support for people displaced by it. Whatever happens in her race, the framing is the one progressives should steal: AI anxiety is real, and the answer to it is a concrete offer, not a flinch.

What you can do

If you work in or near electoral politics, treat AI as a kitchen-table economic issue, not a tech-policy niche. "Here is our plan to make sure AI works for you" is a message that travels. Test it on your own list before a campaign has to guess.

Progressive AI Win

AI is getting people into court who never could before.

The most hopeful AI story this week is buried in a piece about overwhelmed judges. Across the federal courts, people representing themselves, because they cannot afford a lawyer and their case is too small to attract one, are increasingly using AI to draft their filings. Judges are reading stacks of these. The coverage frames it as a problem: the quality is uneven, the volume is heavy. Look again, though. For the first time, people without money or connections are getting into the federal court system in real numbers, and AI is what opened the door. Set that next to the Number of the Week: three in four low-income households face a legal problem every year, almost always alone. This is the first tool that has actually moved that number. It is messy, sure. You fix messy by helping people do it better, not by shutting the door. AI may turn out to be the biggest expansion of legal access since the public defender. Progressives should be naming that out loud and building the scaffolding (vetted prompts, plain-language guides, legal-aid partnerships) before the legal establishment decides the fix is to shut it down.

Put AI to Work

Practical ways progressives can use AI this week

Turn the MIT risk study into a one-page profile for your own community

The MIT study mapped AI risk at the level of 272 global experts. Your org's superpower is mapping it one level down, where it actually lands: on your members, your neighborhood, your constituency. City councils and school boards do not act on global expert surveys. They act on "here is what is happening to people here." Here is how to build that in an afternoon.

Gather a little real texture. Put up a short form (Google Forms or Airtable, ten questions, fifteen minutes to build) asking your people where AI already touches their lives. Does your employer use automated scheduling? Have you gotten an AI-generated denial for benefits, housing, or credit? Does your kid's school use AI grading or monitoring? Collect twenty to fifty responses over a couple of days. You are not after statistical significance. You are after specifics.

Let AI do the analysis. Open Claude or ChatGPT and give it the MIT list of 24 risks plus your responses. Prompt it: "Here are responses from [describe your people, for example low-wage retail workers in my county]. Which of these 24 risks are concretely showing up in their experiences? Rank them by how often and how seriously they appear, and flag any harm here that the MIT list misses."

Turn it into a one-pager a council member would actually read. Ask it to draft a one-page Community Risk Profile: your top three local risks, one real example from your intake for each, and one concrete policy ask per risk. Plain language, no jargon, written for someone with five minutes.

Then a human carries it. AI did the sorting. You bring the one-pager to your next public comment, coalition meeting, or legislative visit. The MIT study made the asymmetry visible to experts. You just made it visible, and actionable, for the people who can do something about it locally.

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

Two threads to keep an eye on.

The Great American AI Act is a discussion draft, which means the preemption language is not final yet. The next few weeks are when public pressure actually matters. Watch which unions and state officials go on record, and whether the three-year freeze survives contact with the people it would silence.

And keep watching the courts. The same AI that is letting people file their own cases is going to force a decision: do courts and legal-aid groups build real support around it, or treat it as a nuisance to be filtered out? The first jurisdiction to officially embrace and guide it will set the template. That is a fight worth showing up for.

Until next time,
Jordan

Know someone who should read this?

Share it with anyone who thinks the only two options on AI are cheerlead or panic.