The Trustee Who Joined the Payroll

Anthropic named its first Chief Global Affairs Officer on 4 August 2026: Mariano-Florentino Cuéllar, former Justice of the Supreme Court of California and until recently president of the Carnegie Endowment for International Peace. Six lines into the company's own announcement sits the sentence that carries the page: "Tino has served as a Trustee of Anthropic's Long-Term Benefit Trust since January 2026. He has stepped down from the Trust to join the company."

The Long-Term Benefit Trust is the structure Anthropic built so that someone who is not on its payroll watches what it does. Five members, a dedicated class of stock that lets the Trust elect and remove a growing share of the board up to a majority, and one explicit rule about independence: trustees may hold no financial interest in Anthropic.

The rule measures what a trustee owns and stays silent on where a trustee goes. Any engineer will recognise the shape of that: the state is constrained and the transition is left open. A trustee forbidden to hold equity may accept an executive post, and the moment he accepts, the prohibition stops applying to him, because the money now arrives by a different route. Independence built this way carries no exit cost.

Four of the five trustees appointed in September 2023 are gone, and Anthropic records each departure itself in a footnote it keeps updating. Jason Matheny left in December 2023 to pre-empt conflicts with RAND. Paul Christiano left in April 2024 for the AI Safety Institute. Kanika Bahl and Zach Robinson both left in January 2026. Cuéllar arrived that same January and left in August. Terms run one year, trustees elect trustees, and failsafe provisions let shareholders alter the Trust's powers without trustee consent where the supermajority is large enough. Every one of those clauses is published.

The calendar is worth a look: the Class T stock was designed to give the Trust a majority of the board within four years of the 2023 Series C, which is roughly now. The body reaches the height of its power with almost every seat held by someone other than the people who designed it.

Cuéllar's own profession solved this problem a long time ago. A judge recuses himself. The courts settled centuries back that confidence in a decision rests on constraints visible to the person subject to it, rather than on the virtue of the person making it. Central banks run cooling-off periods. Market regulators impose windows during which an official may not join a firm he supervised. These are dull, well-tested instruments, and nobody wrote them out of suspicion toward individuals: they exist because a rule resting on character works exactly as well as whoever happens to be standing in it.

Anthropic deserves the concession it has earned. The company disclosed the crossing in its own announcement rather than burying it, and in 2023 it wrote plainly that the Trust was an experiment: "we are empiricists and want to see how it works." No direct competitor has built a body with the power to remove directors. The candour at the outset is real and should be counted.

An experiment produces data, and this one now has some. Three years, four founding members out of five departed, one-year terms, no exit cost, and the first direct crossing to the payroll arriving seven months after an appointment. The data does not say anyone behaved badly. It says the structure offers no resistance in the one direction somebody was eventually going to move.

The pool is small, and that matters more than it looks. Trustees are drawn from think tanks, agencies, university chairs and courts, which is the same pool the executives are drawn from. Membership of a profession does not change when the chair changes; only the counterparty does.

In old Europe this has a name: conflict of interest. Article 68 of the AI Act seats sixty independent experts on its scientific panel, each required to file a publicly available declaration of interests, with the AI Office obliged to manage and prevent conflicts. Commissioners carry a two-year cooling-off period, three for the President. Those rules leak badly: Corporate Europe Observatory counted nine of twenty-six outgoing Commissioners taking corporate roles inside the older eighteen-month window, and the tightening arrived only after Barroso went to Goldman Sachs. A written rule fails in public and its failures can be counted. A rule the watched body writes for itself sets its own exit cost, and sets it at zero.

Anyone who has sat on an internal committee already knows how this ends. The safety group, the ethics panel, the board that signs off releases: you take the seat because it is the seat from which something might be changed, and you learn that the things you might change belong to the same people who decide what you will be doing next year. Voluntary oversight is a role, and roles get resigned. Until leaving one costs something, an oversight seat stays the cheapest promise a company can make.