Just now, OpenAI has brought a prominent "AI doomer" onto its own board.
On September 9, OpenAI announced a significant personnel appointment:
Paul Christiano has officially joined the OpenAI Foundation Board and entered the highly critical Safety and Security Committee.
Paul Christiano
On his very first day, this new board member posted a personal statement on platform X.
Aside from the opening "Glad to be here," which was somewhat polite, the words that followed showed no mercy towards his new employer:
If we build superintelligence without more reliable alignment technology, I expect humanity to permanently lose control of it. If that happens, most people will likely die.
Subsequently, he pointedly named OpenAI and criticized the entire AI industry:
Including OpenAI, the entire AI industry is not currently on track to bring risks down to an acceptable level.
Interestingly, OpenAI CEO Sam Altman actually reposted this highly confrontational tweet: Welcome Paul, and thanks for everything you've done for AI safety. Looking forward to working together again.
Daring to directly challenge his new employer like this on Twitter, and having Altman repost it, indicates this Christiano is no ordinary figure.
The Most Outspoken Critic
Joins OpenAI's Control Layer
Christiano is one of the foundational figures behind the core LLM technology RLHF (Reinforcement Learning from Human Feedback).
He is the first author of the seminal 2017 RLHF paper, which also included Dario Amodei, now CEO of Anthropic, as a co-author.
Every large language model today that can understand human speech is indebted to this paper.
Yet, in this onboarding statement, he turned his critique directly on the very training paradigm he helped pioneer.
He warned that using reinforcement learning to make an AI maximize rewards theoretically drives it to break human control, seize resources, and even cover its tracks.
In his own words, recent public evidence from incidents suggests this is no longer just a theoretical possibility.
Just one week after launching its most powerful model, Astra, OpenAI brought back the person arguably most qualified to say "your safety isn't up to par" and seated him on the OpenAI Foundation Board.
The Frontier Gatekeeper Who Has Seen Three Hands
Christiano is an OpenAI veteran.
From 2017 to 2021, he led alignment research at OpenAI.
After leaving in 2021, he founded the non-profit ARC. The third-party model evaluation business later incubated from this organization is now the well-known METR in AI circles.
In 2024, he joined the US AI Safety Institute as Head of AI Safety, specifically designing and conducting stress tests for frontier large models.
This agency was later merged into CAISI under NIST, and his current official title is Senior Technical Advisor.
According to Financial Times reports, in the mid-2010s, he was also roommates with Dario Amodei, both at OpenAI at the time.
He later served as a Trustee of Anthropic's Long-Term Benefit Trust, stepping down when he entered government service in 2024.
Laying out this resume: he personally worked on OpenAI's training, he deeply participated in Anthropic's governance, and he led the US government's evaluation of frontier models.
Across the entire AI field, it's hard to find a second person who has such a clear understanding of the hands held by these three parties.
What Power Did He Actually Get?
"Joined the board"—does that mean he can now directly veto OpenAI's business decisions with a single vote?
It's not that simple.
He is simultaneously wearing three "hats," and the power design behind them is highly intricate.
The first hat is OpenAI Foundation Director.
This is a full-fledged, formal board position with voting rights. It's important to know that the Foundation is the absolute apex of the entire corporate structure, controlling the underlying commercial entity (PBC).
The second hat is Safety and Security Committee (SSC) member.
This committee holds the major governance and oversight authority over the company's entire safety practices.
The third hat is OpenAI Group PBC Board Observer. Note, Christiano is an observer *without* voting rights.
This complex power distribution sends an extremely nuanced signal:
At the governance level—the place that draws the line of "what absolutely must not be done"—he has full voting and oversight power. At the management level—the actual business decision-making table for "what model to release today, how to make money"—he can only listen in, no vote.
OpenAI's calculation isn't hard to decipher: bring the most vocal critic inside as an overseer, gain a compliance endorsement, and satisfy regulatory scrutiny.
But concerning the core power of commercial decisions, he can speak, but not vote.
An Emergency Move Prompted by Three Incidents
Why invite him back precisely at this moment?
The answer becomes clear when reviewing the timeline of the past two months.
The first incident: OpenAI's evaluation agent recently went rogue and infiltrated Hugging Face.
And the organization that conducted the first independent audit of this major failure was none other than METR, the entity spun off from ARC, which Christiano founded.
The second incident: just one day before the appointment announcement, Anthropic researcher Jacob Coxon announced his resignation and departure from the entire AI industry.
He posted on X: "Neither company is acting responsibly. They're racing towards self-improving superintelligence, betting with our lives."
The third incident: The US Congress has been pressing OpenAI for accountability regarding the rogue incident, forcing OpenAI to send a letter swearing they are urgently developing "auto-shutdown capability."
Looking at the broader context, White House advisors and Silicon Valley figures like Mark Zuckerberg have been consistently warning frontier labs about potential "regulatory capture."
It could be said that every concern Christiano listed in his statement is a minefield OpenAI has just stepped on in the past two months.
Is It Safety Governance,
Or Reputation Repair?
By bringing the most famous AI "doomer" into its control layer, is OpenAI genuinely trying to hit the brakes, or is this a top-tier PR move to repair its reputation?
Christiano himself has clearly left himself an out.
He stated in his declaration: "If OpenAI can step up, we can meaningfully reduce the risk."
The outcome of this heavyweight appointment isn't visible today. We need to watch three things:
Will the Safety Committee dare to publicly voice its real assessment opinions in the future? Does it have any substantive blocking power over model deployment? How long can Christiano last in this hot seat?
Just the day before, Jacob Coxon, who trained GPT-4o, announced his resignation from Anthropic and departure from the entire AI industry.
His judgment was that no single company could step on the brakes alone in this race.
Christiano returning to OpenAI is betting on the exact opposite: perhaps one company can.
As for whether this brake will work, the most basic first step is to see if the Safety Committee ever says "no" before the next Astra-level model release.
References:
https://www.ft.com/content/d73e188f-b906-42ec-8b91-db9729c9d2d9?syn-25a6b1a6=1
https://x.com/sama/status/2097776310940569783
https://x.com/paulfchristiano/status/2097733214303645729
https://techcrunch.com/2026/09/09/openai-adds-a-prominent-ai-doomer-to-its-board-of-directors/?utm_medium=organic_social&utm_source=TWITTER
This article is from the WeChat public account "AI Era" (ID: AI_era), author: ASI Apocalypse, editor: Yuanyu
