What this file is
A follow-up to Anthropic’s AI Pacing Plan — the desk’s extracted critique of Dario Amodei’s “We Must Pace the Frontier” essay (12 Sep 2026). That file already names METR as the example third-party evaluator who would get permanent, employee-level access: desks, badges, laptops, training pipelines.
This file does not invent a METR or ARC Evals donor table. Those filings are not in Nation Runner’s Germany desk source set (Bundestag Grosspenden, the federal lobby register, LobbyControl, DonationWatch, Finanzwende). If a number is not on the first case or the desk, it is not published here.
The rabbit hole is the one the first case already opened: who picks the evaluators, and who benefits if badges become a licensing layer.
What the pacing file already said
Amodei’s step 1 is embedded evaluators. He names METR as an example. Anthropic says it is doing this unilaterally. The desk’s critique is not “never audit labs.” It is that the only binding piece is also the convenient one.
From the case file:
Who picks them? METR and similar groups are not the public. They are a small professional class with its own threat models, funders, and status games. Tim Sweeney's question is the right one: political operatives in lab clothing. Once government starts requiring those desks, you have created a licensing layer. Licensing layers always end up protecting incumbents more than they constrain them.
And:
“Democratic coordination” plus an antitrust waiver is a cartel with a safety brochure. … Challengers, open-weight labs, and anyone training just below the cutoff eat the compliance cost or stay boxed out.
The first file also argues that a slowdown that only binds English-language blog-post labs is slower American AI, not safer AI — because China, distillation, and open weights are not a footnote. That argument does not need a fake grant spreadsheet. It needs the question of whose threat model sits inside the lab.
Why a money map would matter — and why this desk will not fake one
On the Germany desk, Money is scored from disclosed pipes: Grosspenden notices, lobby-register spend bands, revolving-door jobs. The method case is Grosspenden & the lobby register. Access is not a verdict; the receipts are public.
METR-as-badge-holder is a different jurisdiction. The pacing file says the group has funders and status games. It does not name those funders. Nation Runner has not independently mapped METR’s or ARC Evals’ books from public filings. Publishing a “money map” of named foundations, dollar amounts, or board seats that are not in the extract would be invented reporting.
So the honest map, today:
1. Named on the desk: METR is the example evaluator in Amodei’s plan; Anthropic is granting that class of outsider employee-level access; the desk’s critique is capture and incumbency, not “safety is fake.” 2. Asked, not answered: who funds that professional class, who appoints it, and whether a government-required desk becomes a licence that only current frontier firms can afford. 3. Out of scope until sourced: a donor table for METR, ARC Evals, or sibling eval shops. ARC Evals is not named in the extracted pacing file. It is not added here as a fact.
If later public filings land on this desk the way Grosspenden do, they go on a dated update — not in a footnote we guessed.
Who benefits if you never ask
The first case’s private-interest story is: current frontier firms define the speed limit, then ask Washington to make defection illegal. Embedded evaluators with badges sound like transparency. They are also a priesthood that helps decide what counts as an incident and when a model is aligned enough to ship.
A pause-and-audit architecture is easiest to impose on US closed labs, hardest on authoritarian or open ecosystems, and most useful to whoever already has the models, the lobbyists, and the evaluator Rolodex. That sentence is the money map we actually have: incentives, not a 990 we did not read.
Related: Dario Amodei, 12 Sep 2026 — https://x.com/darioamodei/status/2098773920774074715