The Advisory Council

The Atlantic's account of the Anthropic–OpenAI rivalry contains a passage worth sitting with. Daniela Amodei, it reports, worked through conflicts with the help of a large collection of stuffed animals treated as a personal advisory council. Jinji, a cat, was lazy and did not want to do much work. Maura, a pink bear, was a detail-oriented, type-A manager. A panda named Beary Bonds was the patron bear of generosity and empathy. She described the exercises as something between an inside joke and an earnest management technique, and called the animals very useful in analyzing different personality types. The colleagues who knew about the council were not sure how seriously she took its advice.

Let us be fair first. Every executive role-plays; every boardroom runs on personas and rehearsed objections. A leader who names her inner voices and gives them faces is, in one sense, doing honestly what others do invisibly. There is nothing contemptible in it, and this essay is not an attack on a person. It is a witness of a structure.

The structure is this: the council is appointed by the person it advises, its members are voiceless except through her, and its verdicts are whatever she already half-thought, given fur and a name. It is governance improvised in the first person. There is no member of the council who can refuse her — no advisor whose livelihood, reputation, or duty points at the verdict rather than at her comfort. The checkpoint is real in appearance and absent in fact.

This is the cultural wing of the recursive loop. In Stuck in Recursive Loops, the loop was economic: companies pay the evaluators, the evaluators keep the clients, the auditors are partnered with the audited, and nothing outside the circle does the checking. The White House Accord specimen showed the same loop in institutional dress — internal controls, an internal reviewer, an external auditor partnered with the company, a board committee, each layer grading the layer below, every layer paid by the thing being graded. The advisory council is the same architecture one scale down, at the level of a single decision-maker: a loop of consultants who cannot say no, convened precisely so that consultation has occurred.

And it matters where this loop sits. These are the people building the systems that will grade, filter, and advise the rest of us — the people whose companies now write the standards, fund the evaluators, and staff the advisory councils of the republic's technology policy. A culture that settles its hardest conflicts by consulting voices it appointed itself will not notice when the same move is made at scale: alignment evaluated by the aligned, safety certified by the certified, the test written around the wrong problem and graded satisfactory.

The discriminator is the same one as always. A real council contains at least one member who can refuse the convener — whose incentive points at the verdict, not the client. A loop that appoints its own advisors will always find itself well-advised.

The question is never "was advice sought?" It is: who appointed the council, and can anyone on it say no?