Joseph Gordon-Levitt discussing The AI Doc at Sundance 2026. Photo by Jay Dixit.
The view from inside the building
When I was at OpenAI, I didn’t spend much time worrying about AI safety. Not because I thought safety wasn’t important — I was well aware of the staggering risks that come with a technology this powerful. It was that based on everything I was hearing from critics, it seemed like the problem was already being handled.
I led OpenAI’s writing community. My job was to talk to writers, educators, and students who used ChatGPT in the creative process without outsourcing writing or thinking — then share their stories to guide people toward using AI more thoughtfully.
What I heard at the time from critics was mostly misinformed and demonstrably wrong. That AI was all hype. That Sam was promising impossible advances to inflate OpenAI’s valuation. That ChatGPT was nothing but “a bad intern” that would soon go the way of the Roomba. That AI was fast approaching a technological plateau — that because we were supposedly running out of training data, facing “model collapse,” and hitting the limits of scaling laws, the models would therefore never get much better than they already were. Articles predicted OpenAI would run out of capital and go bankrupt — when I already knew we were close to closing the single largest fundraise in history.
Most of the criticism I heard was falsified by what I saw every day. Part of my job was to lead early access cohorts of writers and educators to alpha-test our new unreleased models, so I’d experienced their capabilities firsthand. The critics didn’t know what they were talking about.
Why I wasn’t worried
When it came to safety, the critiques seemed equally unconvincing. It’s not that I thought safety concerns were overblown. Just the opposite: I knew the stakes were existential. But our best people were already on it.
Critics said OpenAI didn’t care about safety, that we were rushing to release models without adequate testing. But I worked in the building. I saw how much time my colleagues spent working to make sure our systems were safe.
These claims came at a time when we’d already held off on releasing Advanced Voice Mode for four months after we’d already announced it. The silence was long enough that users began wondering whether Sam had faked the demo, with an actor hiding off-stage pretending to be ChatGPT. But as embarrassing as the delay was, the safety team refused to sign off until they were confident it wouldn’t be used for voice impersonation or deepfakes — and so we held the launch.
I’d been in the meetings where my colleagues explained how they’d disrupted covert influence campaigns. I’d met the people working on election integrity and child safety. I’d corresponded with red teamers whose job was to stress-test models before release. The company was pouring serious resources into making sure the models were safe.
So when critics said the problem was that OpenAI didn’t take safety seriously, I shrugged it off since I knew it wasn’t true. If the supposed problem was simply that we didn’t care about safety, then there was no problem at all.
The other line of attack I kept hearing was that the world should just stop building AI altogether — which I dismissed just as easily. “Why don’t we just ask the corporations to cease doing the thing they were founded to do?” is not a serious policy proposal.
The critics loved to roll their eyes at that argument and point out how self-serving it was: “So let me get this straight. You have to build this dangerous thing because if you don’t, someone worse will? How convenient.”
And of course it’s true that AI companies also have selfish reasons to want to get there first. Sam Altman once told investors that if OpenAI reaches AGI, it could “maybe capture the light cone of all future value in the universe” — and even Paul Graham called that claim “not totally implausible.”
But just because an argument is self-serving doesn’t mean it’s wrong. Critics refused to accept that two things could be true at once: (1) The CEOs have a financial incentive to make the claim; and (2) despite that incentive, it really is true that the frontier labs can’t afford to slow down because if they do, a worse actor will take the lead.
So in my mind, the safety issue was already being handled. The critics said we didn’t care about safety, but I knew we did. The critics said the world should simply stop building AI, but wishful thinking doesn’t help anyone. AI safety, I concluded, wasn’t something I needed to worry about.
The problem I didn’t see
What changed my mind was seeing The AI Doc, a new documentary about the AI arms race that hit theaters Friday.
The film makes a point I’d somehow never grasped. The problem isn’t that the people building AI are greedy, reckless, and unconcerned about the risks. The problem is that the system itself rewards speed over safety. Good intentions aren’t enough when the rules of the game punish restraint.
The central insight of the film is this: The claim “If I don’t build AGI, someone worse will” isn’t self-serving bullshit like the critics say it is. On the contrary — it’s the whole problem.
It really is true that if the more safety-conscious labs pull back, the lead passes to whoever is more reckless. If American labs slow down, China pulls ahead — which really does make the world less safe. The race dynamic operates lab against lab, nation against nation, and at every level caution is punished.
Slowing down genuinely is dangerous — and that geopolitical reality locks the frontier labs in a winner-take-all race to AGI that incentivizes speed over safety. It’s a profound irony that the labs most committed to safety are the ones least able to slow down — because slowing down would mean handing the lead to someone who doesn’t take the risks as seriously.
Critics like Karen Hao call this argument the “inevitability card” — a convenient excuse the frontier labs serve up to justify charging ahead. But the film shows that dismissing this as self-serving rationalization is the very reason so many critics miss the real problem.
As Connor Leahy puts it: “The worst possible thing that could happen is that there’s multiple AGI projects done all competing to get to AGI first… because this would mean that whoever is the least safe, whoever sacrifices the most on safety to get ahead, will be the person that gets there first.”
In other words, misaligned incentives create a competitive race dynamic that every player hates but no one can escape. “It’s not that ChatGPT is the existential threat,” Tristan Harris explains in the film. “It’s the race to deploy the most powerful, inscrutable, uncontrollable technology under the worst incentives possible — that’s the existential threat.”
Tristan Harris makes the point that the problem isn’t bad intentions, but bad incentives. Photo by Jay Dixit.
Changing the rules of the game
So if no single actor can afford to slow down, does that mean there’s nothing that can be done?
Quite the opposite. AI safety is a coordination problem, which means the solution is to change the incentives for everyone at once. As Aza Raskin puts it in the film: “We need to very clearly change the game from a race to the bottom into a race to the top.”
Aza Raskin speaking at SXSW 2026. Photo by Jay Dixit.
Right now the race goes to whoever moves fastest. A race to the top would mean restructuring the rules such that cutting corners is no longer a competitive advantage. Holding every lab to the same safety standards so no one gets ahead by skipping them.
As Connor Leahy puts it, “There is currently more regulation on selling a sandwich to the public than there is on building potentially world-ending AGI.”
What’s needed, then, is regulation. Not some hippie-dippie appeal to a corporation to please just stop, but changing the game and enforcing rules that apply to every player on the board: requiring labs to disclose what they’re building, meet shared safety standards, submit to independent third-party evaluation, and face legal liability for introducing foreseeable harms.
Even the CEOs are urging governments to impose rules they can’t impose on themselves. Dario Amodei talks often about how it’s up to each of us “to push for the government to get involved.” And Sam Altman has pushed for the creation of an international authority analogous to the International Atomic Energy Agency to coordinate AI safety.
Critics dismiss this as mere posturing: if the companies really cared about safety, they’d simply slow down. But it only looks like hypocrisy if you don’t understand the race dynamic. It’s not a contradiction — it’s an incentive problem.
That’s the piece I didn’t grasp when I was at OpenAI. The labs aren’t lying when they say they want regulation while still racing ahead. The people I worked with really do care about safety. But caring isn’t enough when the race is rigged against caution.
And the pace has only accelerated. Safety testing timelines that once stretched months have now been compressed to days.
The missing piece
So how do we change the rules of the game? The labs want regulation. Legislators can’t act without public demand. The missing piece, then, is public pressure. We citizens have to demand that our governments enable the coordination that no company can initiate on its own.
And since the race operates across borders, that coordination has to be international. That might sound far-fetched, but we’ve achieved it before. At the height of the Cold War, for instance, the United States and the Soviet Union negotiated arms treaties that actually reduced nuclear arsenals.
What most people don’t know is that another film helped make that possible. In 1983, ABC aired The Day After, a made-for-TV drama that depicted the aftermath of a nuclear exchange between the U.S. and the Soviet Union. Almost half of all American adults saw it.
Ronald Reagan wrote in his diary that the film left him “greatly depressed,” and said in his memoir that the film’s depiction of nuclear war drove home a briefing he’d read about “a sequence of events that could lead to the end of civilization as we knew it.”
This was at a time when the Pentagon was still saying nuclear war was “winnable.” The Day After helped Reagan see how crazy that was. Within a few years, he and Gorbachev negotiated an agreement to eliminate entire classes of nuclear missiles.
The Day After helped create a shared understanding of the stakes that made nuclear nonproliferation possible. I’m hoping The AI Doc can do the same thing for AI safety. Coordination becomes possible only when people see the problem clearly enough to demand that it gets solved.
If you want to hear more about writing, creativity, and how to use AI as a sounding board instead of a content generator, please consider subscribing below.
I appreciate you writing this, but as someone who is actually leading a team *fighting for* the regulation you say is needed, it is hard for me to get over the fact that the big Labs and their executives are certainly not ACTING like they want safety rules to solve the collective action problem.
Instead, they’re funding a political operation of unprecedented scale that is singlemindedly opposing ANY safety rules, seeking to overturn existing ones, and wielding hundreds of millions of dollars of super pac spending to intimidate all policymakers from going anywhere near the issue of AI safety. One in four federal lobbyists in Washington DC are currently employed by tech companies to lobby on opposing AI regulation. The forces that have been arrayed to explicitly prevent us from solving this problem are unlike anything I have ever seen, and I don’t think people in tech realize it.
If people in the Labs actually want safety rules that apply universally, then they need to do two things:
1) raise dissent within their companies over the actions of their government affairs teams and political action committees funded by their executives. Challenge the reassuring empty words being said at all-staffs and on Twitter about the leadership’s “good intentions” and contrast it with the concrete actions of the mercenary lobbyists and political operatives they have hired that say the exact opposite.
2) donate to pro-AI Safety organizations like Humans First, which are operating explicitly in the political/legislative advocacy domain. All the technical solutions and white papers don’t matter a jot if nobody will implement them. This is the high leverage, under-exploited area of AI safety.
> But just because an argument is self-serving doesn’t mean it’s wrong. Critics refused to accept that two things could be true at once: (1) The CEOs have a financial incentive to make the claim; and (2) despite that incentive, it really is true that the frontier labs can’t afford to slow down because if they do, a worse actor will take the lead.
Good recognition of the fact that just because an argument is self-serving doesn't mean it's wrong!
I noticed that Karen Hao made this intellectual mistake *twice* in her recent Diary of a CEO podcast interview from last week, (1) when she dismissed the idea that AI poses an existential risk by saying that AI company CEOs profit by promoting this idea, and (2) when she dismissed the idea that scaling the current LLM-based AI paradigm will lead to AGI by saying that AI company CEOs profit by promoting this idea.