Why Seat-at-the-Table Thinking Is the Wrong Frame for AI Safety
Everyone who cares about AI safety wants a seat at the table. I understand the instinct: decisions are consequential, timelines are compressed, and the people making them are few. If you care about the outcome, you want to be in that room.

That’s rational, not naive, but it gets the sequence wrong. In evidence-driven organizations, influence precedes the seat at the table — influence doesn’t come with it.

The case for proximity isn’t wrong – some decisions close before the room empties; informal influence shapes outcomes before reaching any formal process. The right person in the room asks questions that would never get asked otherwise.

Proximity matters.
Pursuing proximity, though, always costs you something: it reorients you toward access rather than the work that forces decision-makers to listen.

Too many people working on AI safety focus on solving the wrong problem. The field attracts people who believe the stakes are high and want to be where decisions get made, a reasonable combination, but one which produces an unreasonable number of people chasing lab adjacency, policy briefings, and government access.

In the right sequence, none of those things are wrong; they’re just downstream of credibility, not upstream.

The organizations doing the most consequential work in AI safety didn’t talk their way into rooms; they produced work so difficult to ignore that the rooms came to them. The mechanism isn’t charisma or timing. It’s that credible work lowers the cost of trusting someone. A decision-maker evaluating a stranger’s judgment in real time takes a risk. A decision-maker who can point to a documented track record is outsourcing that risk to evidence that already exists. Access follows because it’s cheaper to grant than to withhold.

The obvious objection: sometimes people get the seat first and build credibility after — a promotion, a hire, a founder’s title — and it works out fine. That’s real, but it’s not a counterexample to the sequence; it’s a different sequence with different failure modes. Someone who gets the seat before the credibility has to build it in public, under scrutiny, with the cost of being wrong now visible to everyone in the room. That’s a harder version of the same requirement, not an exception to it.

Operational credibility looks nothing like influence-seeking.
Credibility looks like artifacts — documentation, frameworks, a track record that exists independently, whether or not you’re in the room. It looks like the person who identified the gap between what the organization said it was doing and what it was actually doing, and built the system that closed that gap. It looks like prior judgment calls that proved right so frequently that ignoring them became the riskier option. Nobody grants that kind of credibility. You accumulate it; it compounds.

The seat at the table exists; the influence that comes with it is real.

Both are outcomes, not starting points.

People who moved the needle on AI safety didn’t start by pursuing access; they started by producing work that decision-makers couldn’t responsibly ignore, so the rooms came to them.

Don’t seek a seat at the table.
Build a work history that makes the table’s decisions better whether you’re there or not. That’s not a consolation prize, it’s the only strategy that compounds.