First, Do No Harm: AI Under Democracy
As robots take on human abilities, they must be governed by human means. Big Tech is teaching machines to write, reason, diagnose, create, and decide. That means that a handful of billionaire-backed labs are shaping the future of work, of democracy, of our own safety. And they’re doing it all behind closed doors, accountable to no one but their billionaire investors.
One AI CEO, Anthropic’s Dario Amodei, put it plainly: he was “deeply uncomfortable with these decisions being made by a few companies, by a few people.” We agree. Which is why we need a plan to fix it. We created the internet. The American people funded the research that built it through public universities and federal grants. Humanity created the data that populated it. And without it, AI would not exist. AI was built off us. It should work for us, not against us. Americans deserve clarity, control, and a say in how this transformative technology will shape our world.
In January, we released a Data Center Terms of Engagement platform because communities have the right to know and control what’s being built, who benefits, and what they get in return. Now, we’re putting forward a plan to bring AI Under Democracy – guardrails and a roadmap for AI development with one goal in mind: to bring the most powerful technology in human history under human democratic control.
This will ensure that innovation serves all of us who made it possible, not just the few who can profit from it.
1. AI Should be Governed Under Democracy
It’s not a coincidence that Google holds a major stake in Anthropic. Or that Microsoft is embedded in OpenAI. And Amazon holds billions in both. This is a deliberate strategy to control the next platform before it threatens the last one. We have to break up Big Tech’s hold on AI to ensure the public’s best interest is served.
This is not unprecedented. Norway holds board seats in Equinor—it’s massive energy conglomerate—not just stock. American public utility commissions hold veto power over major utility decisions. The Federal Reserve requires public-interest directors who operate independently from the banks. We’ve done this before. Here’s how we do it again:
Mandatory Public Benefit Charters: All frontier AI labs must be chartered as public benefit corporations, legally bound to balance public interest against profit.
“First, Do No Harm” Governance: Charters must enshrine this as an operating standard. Profit cannot come at the cost of public safety, civil rights, or democratic integrity.
Democratic Board Control: 50%+1 of board seats at frontier AI companies must be publicly appointed or democratically elected, not chosen by shareholders.
Mandatory Divestiture: Companies developing frontier AI must operate independently of major platform tech companies. Google, Microsoft, Amazon, and Meta must divest controlling interests within a defined transition period.
No Revolving Doors: Public board members cannot return to employment at any major tech company after their service.
2. Public Ownership, Public Dividend, Public Investments
AI was built on our data and our labor; we should own a piece of it. Senator Bernie Sanders’s American AI Sovereign Wealth Fund Act would transfer 50% equity from the largest AI companies into public hands. I support that effort and propose to take it one step further because Americans deserve a seat at the table.
Warning signs are coming from industry leaders themselves. Geoffrey Hinton, the Nobel Laureate godfather of modern AI, compared this moment to the Industrial Revolution: robots replaced our bodies then, and they’re coming to replace our brains now. Anthropic CEO Dario Amodei predicted that 50% of entry-level white-collar jobs could vanish within five years.
Michigan knows what it’s like to have our jobs and industries ripped from under us, like when profit-driven automation devastated the auto industry in the 1980s. Michiganders can’t afford to go down this road again. This is what public ownership could look like:
We Own Half: Consistent with Sanders’s proposal, with returns flowing to the American people.
Annual AI Dividend: Proceeds fund a direct annual dividend to every American.
Automation Levy: A per-worker-equivalent tax on AI-driven automation, reinvested in workers to do the following:
Education Investment: Education should be free, particular for early and mid-career workers through community college partnerships, guaranteed apprenticeships, and entrepreneurship training.
Expanded Unemployment Benefits and Wage Insurance: Longer duration, higher wage replacement, and a federal bridge that makes up the gap between old and new wages during transition.
Small Business Revolving Loans: Publicly capitalized loans so local businesses can build and thrive.
3. AI Shouldn’t Be Able to Hurt Us
AI is the single most consequential technology humanity has ever created. It has the opportunity to address some of our most pressing challenges, from cancer to climate change. But it also carries the ability to destroy us economically, socially, politically, and physically. And we can’t just sit back and ignore the risks.
We don’t let pharmaceutical companies test their own drugs. We shouldn’t let AI companies test their own models. We have needed guardrails to protect humans from these risks. Here’s what that should look like:
Mandatory Interpretability Standards: It’s not enough to know what the AI said if we can’t understand why it said it. We need mandatory interpretability standards that clarify AI decisionmaking.
Mandatory Behavioral Red-Teaming: Independent adversarial testing for deception, manipulation, and coercion before deployment.
Independent Safety Testing Agency: We need an FDA-style agency that evaluates frontier models before deployment.
Biosecurity Requirements: Mandatory biosecurity red-teaming, coordinated with the CDC, NIH, and FEMA, with restrictions on models that meaningfully assist bioweapons development.
Domestic authoritarianism prohibition: No mass political surveillance, protected-characteristic predictive policing, or AI targeting of dissent.
Self-acceleration Guardrails: AI used to build or improve other AI must face the same safety review as the models it produces. And limits must be placed on the capacity for agents to produce other agents.
Mandatory Incident Reporting: There must be a clear mechanism for reporting AI harms to regulators the way pharmaceutical adverse events are reported.
Democracy Protections: We need a federal ban on deceptive AI-generated political media, mandatory disclosure of AI-generated political ads, and a federal task force on algorithmic disinformation with real enforcement power.
Compute Controls and “Know Your Customer” Requirements: We don’t sell nuclear centrifuges without knowing the buyer. Similarly, chip access must be regulated and traceable on national and international markets.
AI No-Gos: AI may not deny medical care, autonomously fire weapons, conduct warrantless surveillance, replace human oversight in life-or-death decisions, or make hiring or firing decisions.
International Cooperation: We must re-engage the Bletchley framework and global AI Safety Institute network to advocate for a binding compute-threshold treaty with
- Authority to pause dangerous development
- A binding ban on lethal autonomous weapons
- Barring AI any role in nuclear command and control
- Pursuit of a broad multilateral safety dialogue on catastrophic risk, modeled on Cold War nuclear verification cooperation.