In Politico’s new Decoded podcast, Sam said: “We believe that the world should accept some bad things happening for the benefits of this technology and people having the agency.”
The reaction was immediate and hyperbolic. Futurism ran the headline “Sam Altman Says the World Is Going to Have to Suffer for the Glories of AI.” One critic called it “freaking insane.” and “We have to crush these sociopaths.” Zscaler’s CEO went on Yahoo Finance to say Altman is simply wrong.
Here’s my take. Altman is right about the physics, wrong about the conclusion, and sucks at framing the issue. And most of his critics need take some deep breaths before talking.
As always, let’s start with the physics. AI is what lawyers call an unavoidably unsafe product. No matter how carefully you build it, it will cause harm, and some of that harm will be unforeseeable. That isn’t a bug you can engineer away. It’s a property of the thing.
But before you lose your mind, stop and think about it. You already live with lots of these. My chef’s knife is dangerous for exactly the reason it’s useful. A knife that can’t cut my hand can’t cut an onion. Vaccines save millions of lives and, very rarely, seriously injure a child. We don’t ban chef’s knives. We don’t ban vaccines.
But we also didn’t shrug and say “accept some bad things.”
I think the word, “accept,” is what set people off, and they were right to react. “Accept” means the harm lands on whoever it lands on, and they’re on their own. i.e. “It sucks to be you.” That’s not a policy. That’s a shrug. (To be fair, Altman did say catastrophic risks like loss of control should not be accepted. But the everyday harms are the ones that land on real people.)
As I pointed out in a recent blog, we’ve stepped on this rake before. In the 1970s and 80s, vaccine makers lost huge lawsuits, looked at the math, and said, “We’re out.” Congress didn’t pick a side. It built the Vaccine Court. Manufacturers got protection from ruinous liability. Injured people got a no-fault path to compensation from a dedicated fund. Benefits and harms, both on the table, both dealt with.
That’s the mature conversation.
Not “accept the bad things.”
Not “zero bad things or nothing.”
It’s this: the technology will cause harm, so who pays, who checks, and how does the person who gets hurt get made whole?
As part of my fellowship at Harvard’s Berkman Klein Center, I’ve been workshopping an idea: an AI Harms Court modeled on the vaccine program. My first draft folded like a cheap lawn chair when I ran it through my AI Rosetta Stone debate tool and let the accelerationists, safetyists, and skeptics take turns attacking it. Round after round, it got harder to knock over:
→ Immunity earned by measured, deployed behavior, not paperwork
→ Independent auditors the companies don’t pay directly
→ A split levy on developers and deployers
→ No immunity at all for irreversible harms
→ Firms that skip the standards get full tort exposure
Is it right? I don’t know yet. That’s the point.
Let’s stop arguing about whether AI will cause harm. It will. Let’s start arguing about what we owe the people it harms.
Read the post, try the tool, and put your best idea in the ring:
https://www.jsnover.com/blog/2026/09/22/steelmaning-ai-policy-with-the-ai-rosetta-stone/
#AI #AIPolicy #AISafety #ResponsibleAI