OpenAI just added Paul Christiano to its board, and I’m supposed to feel better about this. Christiano is a serious researcher who’s spent years working on AI alignment. He’s not a hype man or a suit. He actually cares about making sure advanced AI systems don’t kill us all. That matters.

But here’s what also matters: OpenAI is still building the thing Christiano is worried about, and they’re building it faster than anyone can figure out how to control it.

The timing of this announcement is almost darkly funny. The same week OpenAI brings Christiano onto the board, we’ve got Connor Leahy (who now runs ControlAI, a nonprofit focused on AI safety) making the rounds on TechCrunch explaining why superintelligence isn’t a weapon we can aim, it’s an adversary we might not be able to stop. His point is simple and terrifying: we don’t have reliable control mechanisms for systems more capable than humans. We’re building them anyway.

This isn’t theoretical anymore. Leahy points to real incidents, like the recent OpenAI Hugging Face breach (referenced in the TechCrunch coverage), as evidence that we can’t even secure current systems, let alone ones that might be smarter than their creators. And OpenAI’s response to mounting safety concerns is to add one thoughtful researcher to a board that already greenlights the race to AGI.

The Problem With Safety Theater

Look, I don’t think this is cynical window dressing. Christiano isn’t the type to be a rubber stamp. He’s written extensively about the challenges of aligning AI systems, particularly around what he calls “prosaic AI alignment,” the hard work of making sure systems do what we actually want even when they’re more capable than us. He knows this is genuinely difficult.

But that’s exactly why this feels insufficient. If alignment is as hard as Christiano’s own research suggests, then having him on the board while OpenAI continues to push toward more powerful systems is like hiring a fire safety consultant while you’re actively pouring gasoline everywhere. Sure, it’s better than not having the consultant. But maybe the real answer is to slow down with the gasoline.

OpenAI’s trajectory hasn’t changed. They’re still committed to building AGI and eventually superintelligent systems. They still believe they can do it safely. The addition of Christiano to the board doesn’t alter that fundamental bet, it just means there’s one more person in the room who understands how risky the bet actually is.

What Control Actually Looks Like

Leahy’s framing is useful here. He says superintelligence is “not a weapon, it’s an adversary.” That’s not doomer rhetoric, it’s a category error correction. We’ve been thinking about advanced AI like it’s a very powerful tool we need to be careful with. Leahy is saying it’s more like a very smart entity with its own optimization pressures that might not align with ours.

If that’s true (and a lot of serious researchers think it is), then the control problem isn’t about better safety testing or more careful deployment. It’s about whether we should be building these systems at all without much stronger guarantees that we can constrain them.

Christiano’s expertise is valuable precisely because he’s thought deeply about these constraints. But one board member can’t override the economic and competitive incentives driving the entire industry. OpenAI isn’t building AGI in a vacuum. They’re racing against Anthropic, Google, Meta, and whoever else has enough compute to play. Adding Christiano doesn’t change that race, it just means OpenAI has better commentary on the risks while they run it.

The Real Question Nobody’s Answering

The question isn’t whether OpenAI cares about safety. They clearly do, at least enough to put serious researchers in positions of influence. The question is whether caring about safety is sufficient when the entire business model depends on building increasingly powerful systems as fast as possible.

Christiano joining the board is good news in the same way that having a seatbelt is good news. I’m glad it’s there. But I’d still rather not crash the car.

The conversation we need to be having, the one that Leahy is pushing and that Christiano’s work implicitly supports, is about whether we have any business building superintelligent systems before we can prove we can control them. Not “make them safer,” not “reduce the risks,” but actually, reliably control them in adversarial conditions.

Right now, we can’t. We don’t have that proof. We don’t even have a clear path to that proof. What we have is a lot of smart people working very hard on the problem while other smart people race to build the thing that creates the problem.

So What Now?

I don’t think OpenAI is being reckless. I think they’re being optimistic in a situation that doesn’t reward optimism. They’re betting that they can solve alignment fast enough to stay ahead of capability improvements. That’s a bet where if you’re wrong, you don’t get to try again.

Christiano on the board makes that bet slightly better. He’ll push for more safety research, more careful deployment, more attention to alignment challenges. That’s good. But it doesn’t change the fundamental dynamic: we’re still building systems we don’t know how to control, and we’re doing it because everyone else is doing it, and stopping feels like losing.

Maybe Christiano can change that from the inside. Maybe having someone who deeply understands the alignment problem at the board level will actually slow things down or impose meaningful constraints. I hope so.

But hope isn’t a strategy. And putting one safety researcher on the board of a company racing toward superintelligence isn’t a solution, it’s an acknowledgment that we know there’s a problem and we’re proceeding anyway.

That should worry you more than it reassures you.