One Advisor, Two Jobs, One New Recusal
OpenAI appointed Paul Christiano to its Foundation Board on September 9, 2026, naming him a non-voting observer on the OpenAI Group PBC board and a member of the Foundation's Safety and Security Committee alongside its chair, Zico Kolter. It's a notable hire on its own merits: Christiano led alignment research at OpenAI from 2017 to 2021, did foundational work on reinforcement learning from human feedback, and went on to found the Alignment Research Center. Bret Taylor, chair of both OpenAI boards, called him "an independent voice on whether the industry's safeguards are adequate" in the announcement.
That independence is also his day job. Christiano currently serves as a Senior Technical Advisor at the Center for AI Standards and Innovation (CAISI), the National Institute of Standards and Technology unit that evaluates frontier AI models for capabilities with national security implications - a role he has held since CAISI's predecessor, the US AI Safety Institute. OpenAI's own announcement carries a footnote most of the coverage skipped past: as Senior Technical Advisor, Christiano "will recuse himself from all OpenAI-related matters as well as all model evaluations." Read that plainly and it says something specific: a government evaluator whose job includes assessing frontier models has just joined the oversight board of one of the two or three labs whose models he would otherwise be assessing, and the fix isn't separation, it's a recusal he has to remember to invoke.
Two Governments, Two Answers To the Same Conflict
The EU AI Act solves the identical structural problem by removing the option Christiano was just given: sit on the inside, then step out when a conflict actually arises. Article 68 of the Act establishes a scientific panel of up to 60 independent experts to support the European AI Office's enforcement work - alerting it to emerging risks, helping build evaluation methods, advising on how to classify general-purpose models. Eligibility is the whole mechanism: every expert must demonstrate independence from any provider of AI systems or general-purpose AI models before they're ever appointed, not after a conflict shows up on the calendar. Each one files a public declaration of interests, and the AI Office runs its own standing process to manage conflicts institutionally, rather than leaving it to any one expert's memory.
| Mechanism | Who it covers | How independence is enforced | What happens on a conflict |
|---|---|---|---|
| OpenAI Foundation Board | A sitting CAISI/NIST advisor, now also a board observer | Not a board eligibility requirement | Named recusal from OpenAI matters and model evaluations |
| EU AI Act, Article 68 | Up to 60 experts advising the AI Office | Independence from any AI provider is a condition of appointment | Public interest declaration; conflicts managed by the AI Office itself |
Neither structure is inherently indefensible. A recusal that's actually honored is real protection, and a seat on a frontier lab's own safety committee gives a skeptic access no outside panel gets. But they are not the same answer to the same question, and only one of them removes the conflict before it can occur rather than after.
What a Compliance Team Should Actually Watch
For a European business tracking either lab as a vendor, the useful signal here isn't the personnel story most outlets ran with - TechCrunch's own headline called Christiano a "prominent AI doomer," which is a personality frame, not a governance one. The useful signal is that OpenAI just picked recusal-based oversight in the same week Anthropic lost two safety staffers outright, including alignment lead Evan Hubinger, who told colleagues he puts the odds of AI causing human extinction within a decade above ten percent on his way out. One lab is keeping its internal critics inside the room with a recusal clause; the other is watching them leave. Both outcomes are visible, both are now part of the public record on how each company actually handles disagreement about its own safety margins - and that record is exactly what a systemic-risk model provider under the AI Act is supposed to be building anyway, whether or not either lab is thinking about Brussels when it makes the call.
Servola Journal
We do this for everyone trying to keep up with what technology is doing to our lives. The people who build it, and the people it happens to. The Servola Journal exists so that what we learn belongs to all of them.
Nobody pays us for this. No ads, no paywall, free to everyone. We just believe that understanding what's happening to all of us shouldn't depend on who can afford to pay for it.
If it gave you something today, tell us to keep going. Follow us, leave a like, or write a positive comment. We read every one, and they are what keeps us going.
Read next: The Only Binding AI Check Left Is European | Anthropic's Alignment Lead Put a Number on AI Risk
