Fear Itself: AI Anxiety Is No Excuse for Self-Regulation

The stakes of AI’s breakneck, barely regulated explosion have never been clearer than they have this past week.

Anthropic researcher Jacob Coxon resigned, warning that the people building AI privately believe it could kill everyone within the next decade. Colleagues who stayed, including the scientist whose job is to make sure the company’s AI does what it’s supposed to, backed the warning publicly, citing incredible odds of doom. Days later, Anthropic CEO Dario Amodei, citing in part the sci-fi-esque recent incident in which OpenAI AI agents hacked the company Hugging Face, argued for a set of voluntary measures by labs, individually and collectively, to slow the pace of development. Elon Musk and OpenAI CEO Sam Altman quickly endorsed the idea. 

There are good reasons to approach the companies’ safety claims with skepticism. One is to worry that focusing too much on existential risk may obscure the more concrete and immediate societal harms we can anticipate. If we have limited bandwidth for attention and reform, we can’t leave those concerns behind. A second is that companies can use “safety” as a moat: The scarier the risk sounds, the easier it becomes to justify barriers that also lock in whoever’s already at the frontier, and leave everyone else permanently behind. These are real and valid concerns. 

But we don’t have to buy all of the most outlandish company claims to conclude that there are real dangers to contend with here. In this same week, Anthropic’s own threat intelligence report documented state-sponsored groups, financially motivated criminals, and propaganda operators actively weaponizing Claude for cyberattacks, influence campaigns, and weapons development. Whatever we think of the self-interest of the companies, there are real risks to be guarded against.

The companies are asking for room to coordinate among themselves. They’re asking for antitrust exemptions, allowing them to set up self-regulatory regimes. These are some of the same companies and their investors donating massive amounts of money to defeat pro-regulatory electoral candidates, wielding their economic power in the political sphere to ensure lock-in of a self-regulation-only regime. 

Here’s where I think the real contest lies. We cannot allow self-regulation to be the path we take. Yes, this conversation is about risk. But more than anything, it’s about power and control—who decides what risks we take: us or them. 

If coordination is necessary for safety, then the federal government needs to make sure the public has a seat at that table, through whatever means possible. The Defense Protection Act has one such mechanism for allowing coordination with affirmative oversight and engagement from the government and other interested parties like labor unions. 

But those existing authorities won’t be sufficient to drive control over the direction that technological development will take, and we’ll need more bold proposals like Sen. Sanders’s, which would take a 50 percent public equity stake in these companies and give the American people a bigger say in all of these decisions.

Industry will argue that government is too slow, or doesn’t have the technical expertise to adequately stand up a regulatory regime. It wouldn’t be credible to deny these concerns entirely; our government institutions have been sclerotic and unfit for purpose for decades. But these are also excuses to lock in self-regulation, and we have to respond with ambition rather than resignation.

Perhaps more disturbingly, we have to contend with the fact that this administration has a demonstrated lack of interest in taking regulation seriously (President Trump said so on Sunday) and, to the degree it does take an interest, it’s one of self-enrichment. What do we do when we know people should have a seat at the table but their elected representatives are compromised?

The answer isn’t an absence of democratic governance. We have to look to other sites of power. 

We don’t need to settle the debate over exactly how dangerous AI is, or prove our regulators are fully up to the task, before acting. Now is the time to protect workers, consumers, and citizens by putting us rather than the AI companies in the driver’s seat.

First of all, we live in a federalized system for a reason; the states—governors, attorneys general, state legislators—provide an alternate source of countervailing power to the labs, and ones that have proven their greater inclination and ability to act. Now is the time to get creative on how those actors can intervene individually or collectively to help craft the right relationship between the labs and the public’s representatives.

And second, it has always been true that the real defense against coercive power in both the private and the public sphere has been civil society. The backlash against data centers is the clearest evidence that the public can and will organize in this moment. What we need is a ‘no’ coalition large enough to demand real power and say in what an affirmative regime should look like: one that has labor and civil society at the table crafting the future.

Solving ‘AI safety’ is today’s crisis, but the deeper task is bigger than that: locking in public, democratic control over this technology rather than private self-regulation. What we do today will determine who’s in charge for decades. 

We don’t need to settle the debate over exactly how dangerous AI is, or prove our regulators are fully up to the task, before acting. Now is the time to protect workers, consumers, and citizens by putting us rather than the AI companies in the driver’s seat.