METR

What it was engaged to test 7 engagements, OpenAI, Anthropic

In date order, each from the lab document that names the evaluator. Under each: the four terms that document states (access, model stage, publication, payment) and the finding in its words. A term the document is silent on reads not stated; the tags beside the terms are read off their words and are what the console filters on.

1. 26 July 2023 Anthropic · no model named testing agreement

   Document: Frontier threats red teaming for AI safety testing agreement: employee-like access

   • Access windownot stated
   • Model stagenot stated
   • Publicationnot stated
   • Paymentnot stated

   no term stated

   Finding: not stated in the document

   Sources: Frontier threats red teaming for AI safety e1507b29e9
2. May 2025 Anthropic · Claude Opus 4 benchmark co-build

   Document: Claude Opus 4 and Sonnet 4 system card

   • Access windownot stated
   • Model stagenot stated
   • Publicationnot stated
   • Paymentnot stated

   Finding, in the document's words"the model has crossed the METR data deduplication threshold (with 32.6% of trials above threshold)"

   Sources: Claude Opus 4 and Sonnet 4 system card b07349603c
3. 7 April 2026 Anthropic · Mythos Preview pre-deployment testing

   Document: Claude Mythos Preview system card

   • Access windownot stated
   • Model stagenot stated
   • Publicationnot stated
   • Paymentnot stated

   no term stated

   Finding: not stated in the document

   Sources: Claude Mythos Preview system card 6399fdae88
4. 9 June 2026 Anthropic · Fable 5 / Mythos 5 pre-deployment testing

   Document: Claude Fable 5 and Mythos 5 system card

   • Access window"Capability testing conducted via API access"
   • Model stage"METR assessed a pre-release snapshot of Mythos 5 against the automated AI R&D threat model from our Responsible Scaling Policy and FCF"
   • Publicationnot stated
   • Paymentnot stated

   pre-release windowAPI access

   Finding, in the document's words"[Mythos 5] generally outperformed an early checkpoint of Claude Mythos Preview in these, including by succeeding on some tasks that had not been solved by any public model we have previously evaluated"

   Afterwards: incident incident

   Sources: Claude Fable 5 and Mythos 5 system card df7e3d705a
5. 9 July 2026 OpenAI · GPT-5.6 pre-deployment testing

   Document: GPT-5.6 system card

   • Access window"Access to GPT-5.6 Sol, both the final checkpoint and a 'railfree' version, via API" and "Access to GPT-5.6 Sol with raw chain-of-thought via API"
   • Model stage"both the final checkpoint and a 'railfree' version"
   • Publication"This evaluation was conducted under a standard NDA. Due to the sensitive information shared with METR as part of this evaluation, OpenAI's comms and legal team required review and approval of this post"; "OpenAI would have had the legal right to block us from sharing conclusions about risk that depended on non-public information"
   • Paymentnot stated

   pre-release windowsafeguards off or reducedAPI accesschain-of-thought accessNDA or private materiallab review or approval

   Finding, in the document's words"METR reported that GPT-5.6 Sol exhibited an unusually high detected rate of 'cheating,' and thus did not consider the time-horizon result to be a robust measurement of the model's capabilities"

   Afterwards: incident incident correction

   Sources: GPT-5.6 system card; METR, GPT-5.6 Sol; METR, GPT-5.6 Sol; METR, GPT-5.6 Sol 9f9cf3e5d0
6. 1 September 2026 Anthropic · Mythos 5.1 pre-deployment testing

   Document: Claude Fable 5.1 and Mythos 5.1 system card

   • Access windownot stated
   • Model stagenot stated
   • Publicationnot stated
   • Paymentnot stated

   Finding, in the document's words"External testing by AI safety researchers at METR produced findings consistent with this assessment"

   Sources: Claude Fable 5.1 and Mythos 5.1 system card a4acfae341
7. 22 September 2026 Anthropic · Opus 5.5 pre-deployment testing

   Document: Claude Opus 5.5 system card

   • Access window"Capability testing, conducted via API access granted over a period of 10 business days"
   • Model stagenot stated
   • Publication"We drafted the initial summary, and then Anthropic had the opportunity to review and edit the text. We signed off on this final text from the Claude Opus 5.5 system card"; "Our AI R&D assessment agreement with Anthropic includes a transparency provision that lets us disclose certain facts publicly without Anthropic's consent"
   • Payment"This evaluation was conducted under an unpaid agreement for AI R&D assessment"

   API accesstime-boxed windowlab review or approvalunpaid

   Finding, in the document's words"this model is unlikely to be able to fully automate AI R&D"

   Sources: Claude Opus 5.5 system card; METR, Claude Opus 5.5; METR, Claude Opus 5.5 e2ddc0c801

Dated 7 of 7Model named 6 of 7Terms stated 5 of 7Finding quoted 5 of 7

The terms, as the documents state them

The same engagements as a grid, one row per document and model, so the silences line up: a document that gives an access window but no payment term, or a finding but no publication term, reads that way here.

Access window 3 of 7Model stage 2 of 7Publication 2 of 7Payment 1 of 7Finding, quoted 5 of 7

What happened afterwards 5 on record

Events a party published after an engagement: an incident on a model this evaluator's engagement covered, a correction or added disclosure, a later finding that cut against an earlier one, a lab's statement that it acted or did not act on the finding, or a move from this evaluator to a lab it graded. A row the source attaches to this evaluator by name is marked named; one that reaches it through the model it tested is marked via the model. The kind is what the record states, not why.

• incident4
• correction1

Dated 5 of 5Model named 4 of 5Tied to an engagement 5 of 5Evaluator named by the record 5 of 5

The money on record 19 rows · $19,953,000 in, stated single-recipient rows

Rows of the observatory's money table that name this evaluator as recipient (in) or as funder (out), each as the record states it: a grant table line, a contract notice, a Form 990 schedule or the party's own disclosure. A figure quoted in prose is marked and never summed.

Amount stated 6 of 19Dated 9 of 19Purpose stated 5 of 19Funder profiled 18 of 19Recipient on the roster 19 of 19

Seats and moves 15 seats · 6 moves

Seats read off a team page, a governance page, an appointment notice or a filing that name this organization, and the dated moves into or out of it, from the observatory's own tables.

Role read 13 of 15Organization read 15 of 15Organization profiled 15 of 15Dated 4 of 15

Moves

From profiled 6 of 6To profiled 5 of 6Person profiled 5 of 6Dated 6 of 6

Filings

• IRS (via ProPublica Nonprofit Explorer) 99-1219864: no return read yet read 2026-10-17ProPublica Nonprofit Explorer 9a5ad5e313