Three years ago Head of Safety at a foundation-model lab was a research seat answering to a chief scientist. Today, at the labs we work with, it is a research-and-policy seat answering to the board through a chief executive, with a standing item on the board agenda and a quarterly report to a subcommittee. The seat has moved. Most hiring briefs we see for it have not.
What boards typically ask in the brief
A research record, ideally adjacent to alignment work. Publication history. The ability to lead a team of senior researchers. A view on the major open questions. These are sensible things to ask for; they are not, on their own, the things the seat now requires.
What the seat now requires
Three additional capabilities the standard brief omits. The first is the ability to brief a board, in lay terms, on technical risk, and to do so under conditions where the board has read coverage in the FT that morning and arrives with prepared questions. This is not a research skill. It is a closely-related-but-different skill that researchers have varying capacity for and that the brief almost never tests for.
The second is the ability to negotiate with policy teams — inside the firm and outside it — about which technical claims can be made publicly, with what hedging, in what venue. This is a craft and it is the craft most often missing on the senior bench.
The third is the ability to make and own a decision to slow or stop a deployment, in opposition to commercial pressure, in front of an audience of named executives. This is not about technical correctness. It is about institutional weight — the candidate has to be the kind of person whose decision sticks. Hiring for institutional weight is something boards know how to do in finance and almost never know how to do in AI.
“The candidate has to be the kind of person whose decision sticks. Boards know how to hire for that in finance. They are still learning to hire for it in AI.”
How the wrong hires look from outside
Two patterns. The senior researcher who is technically excellent and politically thin: they are right about the issue and ignored about the decision. The senior policy leader who is politically expert and technically thin: they make the right institutional moves on the wrong questions. Both patterns are visible in public commentary about the major labs over the last eighteen months. Boards have not generally adjusted their next brief in response.
What the rewritten brief looks like
Three additions to the standard requirements. First, an interview specifically on board communication — the candidate is asked to walk a non-technical audience through a hard call they made and were challenged on. Second, references taken not only from research peers but from at least two non-research executives the candidate has worked with closely; the question to them is whether the candidate's decisions stuck. Third, an explicit conversation about the conditions under which the candidate would resign. This is the conversation boards skip. It is the most important one.
The hiring decision is also an institutional decision
Hiring a Head of Safety in 2026 is not analogous to hiring a Head of Research. It is closer to hiring a Chief Risk Officer at a bank in 2010. The firms that treat it as a research hire produce safety leadership that is technically sharp and institutionally light. The firms that treat it as a senior executive hire — with the rewritten brief, the cross-functional interview, the harder reference process — produce safety leadership that holds. The latter outcome is what the seat is now there to deliver.
