Anthropic → METR
Held pilot review-publication veto
2026-09-16
METR says Anthropic had a general veto over publication of its February 2026 risk-report review under their pilot NDA.
Connects evaluation, philanthropic support and public decision-making
METR
METR develops AI evaluations and helps turn their results into decisions about development and deployment.
METR occupies the space between building a model and deciding what to do with it. Technical assessments can become inputs to company release decisions, public risk debates and government policy. That makes the terms of evaluation as important to this investigation as the headline test results.
Its funding and relationships now have their own map: Canary with RAND, philanthropic commitments, government work and access supplied by developers. The useful questions concern who selects the tests, who can see the results and who can challenge a consequential assessment.
METR says it has not accepted frontier-company funding and does not accept donations made by or at the direction of their staff, while accepting substantial free tokens. Financial separation coexists with practical access dependence. Access agreements and publication terms would show how that tension is managed.
METR: partnerships and funding ↗METR funding update ↗New Support Through The Audacious Project ↗Canary: Evaluating Frontier AI ↗Responsible Scaling Policies ↗
Assessment updated 2026-09-14
Detailed records and research
What the records show
Ajeya Cotra → METR
Staff researcher and investigation coauthor
2026-08-26
METR's August 26 report identifies Ajeya Cotra as a staff member and coauthor of its investigation.
METR → Responsible Scaling Policies — framework
Prototypes scaling policy
Event or period: 2023-09-26; checked 14 September 2026
ARC Evals, METR’s predecessor, presented voluntary commitments as a testbed for regulation.
METR’s August 2026 funding commitments → METR
Approximately $71m committed
Event or period: 2026-08-14; checked 14 September 2026
Raised over preceding six months; donor shares unspecified.
European AI Office → METR
Technical-assistance contract
Checked 14 September 2026; event date unresolved
Supports methods for assessing loss-of-control risks.
Joe Benton → METR
Plans to join
Checked 14 September 2026; event date unresolved
Biography says he will soon join; start unconfirmed.
Canary — RAND–METR collaboration → METR
Approximately $17m for METR
Event or period: 2024-10-09; checked 14 September 2026
Part of the $38m package, not additional funding.
Anthropic → METR
Provides access and tokens
Checked 14 September 2026; event date unresolved
METR acknowledges in-kind evaluation resources.
From the investigation
How research enters the political argument
Sanders’s September 3 essay cites Ajeya Cotra and Dwarkesh Patel to make the case for a pause. METR identifies Cotra as staff and a coauthor of its August incident investigation; it says OpenAI did not pay for that assessment. The essay footer credits Friends of Bernie Sanders. These are observable steps in communication: research, public interpretation, citation by a politician and distribution. They do not establish that the cited people endorse every proposed penalty, or independently verify the incident described.
Read the research & sources ↗METR and Anthropic: the actual levers
METR reports about $71m in recent commitments , says it accepts no frontier-lab funding or staff-directed donations, and acknowledges substantial free model tokens. It lists Anthropic among five labs providing evaluation access and tokens. Its donor list does not establish an Anthropic cash grant. Funding update · METR About .
Read the research & sources ↗METR and Anthropic: the actual levers
The sharper document is METR's review of Anthropic's February 2026 risk report . METR criticized the report's evidence and survey while agreeing, based partly on outside evidence, with its very-low-risk conclusion for Opus 4.6 R&D automation. In its process appendix, METR says Anthropic reviewed publication under an NDA and had a general veto during this pilot. METR says no redactions happened, published its criticism, and wants future review limited to sensitive-information redactions. A contractual veto is an independence risk; this record does not show that it was exercised or that METR's conclusion was purchased.
Read the research & sources ↗A funded collaboration with RAND
METR’s October 2024 announcement describes about $38m catalyzed through Audacious for Canary, including about $17m for METR. Gates appears in the wider collective; a Gates-specific Canary contribution is not identified. The project profile describes evaluation capacity as the intended outcome.
Read the research & sources ↗Further reading
See the connections
Open this part of the network ↗Rules & government
METR: funding, access and policy
Follow grants, lab access, people and public-sector work around the evaluator.
Open diagram →Rules & government
Gates, Anthropic and public-service AI
Follow the joint programme and its institutional connections.
Open diagram →What the connections say
13 relationshipsAjeya Cotra → Staff researcher and investigation coauthor → METR
2026-08-26
METR's August 26 report identifies Ajeya Cotra as a staff member and coauthor of its investigation. Primary organizational account; this branch did not evaluate the full technical report. METR says OpenAI did not pay for the assessment.
Joe Benton → Plans to join → METR
Checked 14 September 2026; event date unresolved
Biography says he will soon join; start unconfirmed.
Read the original source 1
Alignment Research Center (ARC) → Incubated before announced spin-off → METR
Event or period: 2023-12-04; checked 14 September 2026
Announced standalone nonprofit separation.
Read the original source 1
Canary — RAND–METR collaboration → Approximately $17m for METR → METR
Event or period: 2024-10-09; checked 14 September 2026
Part of the $38m package, not additional funding.
Read the original source 1
METR’s August 2026 funding commitments → Approximately $71m committed → METR
Event or period: 2026-08-14; checked 14 September 2026
Raised over preceding six months; donor shares unspecified.
Read the original source 1
Schmidt Sciences → Acknowledged supporter → METR
Checked 14 September 2026; event date unresolved
Named among supporters over the years. Amount and payment date unspecified.
Read the original source 1
David and Lucile Packard Foundation → Acknowledged supporter → METR
Checked 14 September 2026; event date unresolved
Named among supporters over the years. Amount and payment date unspecified.
Read the original source 1
Longview Philanthropy → Acknowledged supporter → METR
Checked 14 September 2026; event date unresolved
Named among supporters over the years. Amount and payment date unspecified.
Read the original source 1
UK AI Security Institute → Acknowledged supporter and partner → METR
Checked 14 September 2026; event date unresolved
METR lists AISI support and partnership. Amount and terms unspecified.
Read the original source 1
European AI Office → Technical-assistance contract → METR
Checked 14 September 2026; event date unresolved
Supports methods for assessing loss-of-control risks. Value and terms not recovered.
Read the original source 1
METR → Prototypes scaling policy → Responsible Scaling Policies — framework
Event or period: 2023-09-26; checked 14 September 2026
ARC Evals, METR’s predecessor, presented voluntary commitments as a testbed for regulation.