OpenAI Brings Its Sharpest Critic Into the Boardroom
Last week, Jakub Pachocki, OpenAI’s chief scientist, asked in an essay for a referee: mandatory safety standards, audited by external auditors or governments. What his company did this week reads like a follow-up step. On Wednesday, September 9, 2026, OpenAI announced on its own news page that Paul Christiano is joining the board of the OpenAI Foundation, the foundation board above the company, and the Safety and Security Committee, the committee that oversees safety and security. In OpenAI’s own words, he brings “experience in AI alignment, safety, and standards”: experience in alignment (ensuring that AI does what the creator intends), safety and standards. The committee is led by Carnegie Mellon professor Zico Kolter, TechCrunch reports.
What that polished announcement does not say: Christiano is not just any safety expert joining the team. He is arguably the sharpest critic of the pace at which the sector, OpenAI leading the way, is racing toward ever more capable AI.
The Man Who Says the Sector Is on a Collision Course
TechCrunch called him in the headline of its piece “a prominent AI doomer”. Christiano’s own words, recorded by that same TechCrunch, do not mince matters:
“I now believe there is a meaningful risk that rapid acceleration in AI capabilities leads to catastrophic and irreversible loss of control in the very near term.”
He believes there is a meaningful risk that the rapid acceleration of AI capabilities will lead to catastrophic and irreversible loss of control in the very near term. And he does not spare his new employer: “I do not think that the AI industry in general, including OpenAI, is currently on track to reduce this risk to an acceptable level.” According to him, the industry, OpenAI included, is not on track to reduce that risk to an acceptable level.
This man now sits on the board of the foundation above OpenAI and on the committee that oversees safety. Pachocki asked last week for a referee; his company is now bringing one in-house. The question that lingers: does the brake get a real hand on the wheel, or is OpenAI mainly buying credibility?
Same Wednesday: Accelerating and Braking at Once
Anyone who thinks OpenAI is easing off the gas only needs to look at the rest of that same Wednesday. At 11:00 GMT, the company launched “GPT-6 Astra: The next generation in intelligence for work”, a business rollout of its latest model. At 13:00 GMT came a policy plea from Chris Lehane, “The AI policy window is open. We need to act”, centered on the idea that stronger AI capabilities demand stronger safety evidence, shared standards and sustainable policy. And at 17:00 GMT came Christiano’s appointment (all times from OpenAI’s own news feed). Product, plea and brake, neatly one after another on a single day: the same pattern as the Sunday when the gas pedal and the brake came online an hour apart.
The appointment also falls within the period when oversight of OpenAI suddenly became very concrete: since Monday it has been confirmed that the incident report about the hijacked German wiki is in Brussels. Whether those two are connected, we do not know; OpenAI says nothing about it, and we are not going to speculate. What is clear: the company continues to do what it has been doing for weeks: push and brake at the same time, and announce both loudly.
Such a Person Does Not Sign On to a Fig Leaf
The easiest cynical reading, bringing in a critic as window dressing and changing nothing else, is too easy here. Christiano is not a prop you set down for a photo. He co-developed reinforcement learning from human feedback at OpenAI, in short RLHF, the technique by which a model learns from human feedback and which is part of the foundation of ChatGPT. He left in 2021, founded the Alignment Research Center, an independent research center on AI safety, and today advises the U.S. Center for AI Standards and Innovation. Someone who built his reputation by leaving OpenAI and calling the sector on a collision course does not lightly put his name behind a PR exercise. And a seat on the safety committee is more concrete than an essay: an essay hopes, a committee member sits at the table.
And that committee is no talk shop. According to TechCrunch, it has “the final say on whether OpenAI releases new models”, the final word on whether OpenAI releases new models, like the Astra it launched last week. The brake does indeed have a hand on the wheel. What we do not know is how much weight Christiano’s voice carries within it: about his personal mandate, and about who decides when he thinks it is going too fast, the announcement says nothing. The difference between a hand on the wheel and a voice from the back seat does not lie with the committee, but with his place in it.
So follow not the press releases but the launches. Christiano has been speaking loudly for years about what he thinks is wrong, even to the point of saying his new employer is not on track. If from now on you see launches slow down or fail when the committee hesitates, and he keeps speaking as loudly as he did outside, then the brake got a real hand on the wheel this week. If launches just roll on while things go quiet around him, then that seat was mostly a buy-in on the credibility market.
