
Image by Anthropic via Flickr
Op-Ed by David Munir Nabti – Visiting fellow at UC Law SF's Center for Constitutional Democracy
I've barely slept these past couple of weeks, worried about how decisions by a tiny number of people at a company just across the bay will shape all of our futures.
This week Anthropic, the company behind the AI model Claude, is going public. I worry the IPO will expose it to the whims of short-term traders and push it toward more reckless decisions. And I worry it sets a precedent for other AI labs that care far less about safety.
Admittedly, I carry some trauma here. I'm a Bay Area native who grew up at Stanford, went on the Friday night skates around San Francisco, and studied at UC Berkeley. I've also lived in Lebanon through cycles of collapse, saw the banks lock their doors, the currency lose more than 98% of its value, and our windows get blown out by one of the largest non-nuclear explosions in history. I attribute all of that to failed governance and a lack of democratic accountability.
We're struggling with accountability in the US too, in a country with massive corruption, and more guns than people, with a higher tendency to use them than any other developed country. So when people talk about AI destabilizing the economy, society, or our democracy, I probably have darker visions of what that could mean than most people in tech. I've seen a version of the collapse I hope we can steer away from.
Frontier AI labs are building technology that could be more consequential than nuclear weapons. Yet their governance answers to a handful of people or to short-term traders, not to the larger population who will be impacted by their decisions. So I spent these last weeks deep in Anthropic's S-1 and governance documents, trying to think of how to help guide the company toward steward governance before the IPO. I knew it was a long shot, and as you read this, Anthropic may already be public. But it led to great discussions, connecting with some great people, and great ideas. And through all that, I had an important realization: many of these critical changes that can strengthen steward governance can still happen after the IPO.

Anthropic Co-Founder & CEO Dario Amodei. Photo via Techcrunch ơn Wikipedia
It all comes down to seven people
To be fair, Anthropic has done more than most AI labs to build in safeguards, and that gives us a greater ability to fix the gaps.
One of the biggest issues: Anthropic's seven co-founders together hold the controlling vote on basically every major decision. Even if we assume they always act in the best interests of all of us, it's still just seven people. Accidents happen. Burnout happens. People disagree, change direction, or retire.
As that group dwindles, much of their power passes to shareholders on the public markets. That's where even more risk sets in. In typical corporate governance structures, someone who bought a share yesterday gets the same vote as someone investing for decades, and far more say than the public. With markets reacting to every headline, and some people putting pressure for profit maximization with minimal or no regard for harms done in the process, the pressure for quick wins can come daily. But proper fiduciary duty should be measured in decades and generations, not days.

Anthropic's current corporate structure
A steward council instead
What if, as the Anthropic founders step back, their control passed instead to a long-term steward council holding a controlling share? Founders, employees and other investors could contribute shares, and more investors could pledge to vote alongside it.
This could connect to Anthropic's Long-Term Benefit Trust, but today it doesn't. The Trust's main job is choosing four of the board's seven directors. Important, but limited.
And power shouldn't just pass from one small group to another small group. With technology this consequential, there should be more voices involved, and more shoulders to carry that burden. There could be a large citizens' assembly (50 people? 100? 200?) selected on a rotating basis (perhaps serving part-time for 1-3 years). Or a multi-tier structure like most governments: a large, randomly selected group from many communities, backed by expert staff. Random selection, or sortition, brings in diverse voices, builds consensus across party lines, and guards against corruption, conflicts of interest and ego. Those people should all be compensated, but at a fixed rate, not based on financial returns. Over time the council could hold shares in other AI labs too, or spread the model to others.
The same body could also help direct the IPO's charitable windfall. Rather than those funds getting allocated by a small number of people or sitting with fund managers, more of those billions should help nonprofits, social enterprises, and communities build long-term impact, sustainability, and endowments of their own, a case Glen Galaich makes in his book Control. (A topic of another op-ed.)

Anthropic’s offices in downtown San Francisco. Photo via Wikimedia Commons
One step at a time
We're not asking for this trust outright. Anthropic's own leaders have called their governance an experiment that will evolve, and the ideas here are starting points for further discussion. We want to build these structures together with Anthropic's employees, people at other companies, civic leaders and diverse communities. As we build these foundations and demonstrate effectiveness, shareholders and eventually the founders could hand them more authority, step by step.
Thankfully large steward-owned companies already exist, and so there are great models to build on. Patagonia put its voting stock in a purpose trust. Novo Nordisk is publicly traded while its foundation holds about 77% of the votes. But none face decisions as consequential as Anthropic's, so we'll need to evolve those models to suit our current situation.
Join us
To the Anthropic community: Please join these conversations and help shape what comes next. If you plan to give shares to charity, ask who will vote them. To the civic and deliberation community, and anyone who cares: please also bring your diverse experiences and ideas to help strengthen this movement.
This isn't against Anthropic. It's for Anthropic, for the Bay Area, for the world, and for our children and grandchildren. Let's build a better and brighter future, all together.
You can learn more about the ideas behind the call for a 90-day delay in the ANT IPO here, and our efforts to foster more long-term collaborative stewardship here.
David Munir Nabti is a Bay Area native, Chief Steward of DemocracyWeek.net, and a visiting fellow on AI and democracy at the Center for Constitutional Democracy at UC Law San Francisco. He writes only for himself, not for the Center or UC Law SF. He's sparking a series of discussions with people at Anthropic and other frontier AI labs about how we can design and build these impact efforts to strengthen our democratic decision-making and future. If you're interested in joining those public or private discussions, please see more info at https://democracyweek.net/stewards.







