SecureBio
What it was engaged to test, on what terms, what it found in the document's own words, and what happened afterwards. Every line is a record; every record links to the document it was read from.
▣ THE RECORD
nonprofit
Bio contractors
STORE 2026-10-23
- Foundednot stated
- HQCambridge, Massachusetts
- Registry idEIN 88-3068255
- TestsVirology knowledge (VCT).
- Money, as the roster puts itNonprofit; Good Ventures $1,630,000 paid, tax year to 30 Jun 2025.
- Filesdossier · observatory, sliced to it · graph
- Models
- Claude Opus 4Fable 5 / Mythos 5Fable 5.1 / Mythos 5.1GPT-5.5GPT-5.6GPT-6 AstraHaiku 5.5Muse SparkMythos PreviewOpus 4.6Opus 4.8Opus 5Opus 5.5Sonnet 5.5
- Span of engagements
- May 2025 – 7 October 2026
- Terms and findings
- 4 of 14 documents state a term · 4 quote a finding
- Money in, US dollars, stated single-recipient rows
- $33,187,487 19 in · 0 outfrom Coefficient Giving · European AI Office · Frontier Model Forum AI Safety Fund · Good Ventures Foundation · OpenAI Foundation · Sentinel Bio · Survival and Flourishing Fund
- Seats, by side
- evaluator 2 between 1 3 people: Dan Hendrycks, Kevin Esvelt, Sawyer Bernath
- Afterwards
- incident 2
Each count is the number of store records that name this evaluator. An engagement is one lab document naming it for one model; an outcome is an event on the record afterwards; the money, seats and moves are the observatory's rows filtered to it. A cell that cannot be filled says not stated, and each table ends with how many of its rows state each field.
▣ SLICE
39 / 39 records
Filters combine: any of the values ticked within a dimension, all of the dimensions at once. A table that has no column for a dimension you have set says so instead of showing rows that cannot match. The address bar carries the slice.
SLICE
edit ↑
What it was engaged to test 14 engagements, OpenAI, Anthropic, Meta
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.
- 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"Claude Opus 4 does substantially better in the end-to-end evaluation of both virology tasks compared to Claude Sonnet 3.7, achieving a 'max of 5' score of 0.83 for virology task 1, and a 'pass@5' score of 0.720 for virology task 2"
- Document: Claude Opus 4.6 system card
- Access windownot stated
- Model stagenot stated
- Publicationnot stated
- Paymentnot stated
Finding: not stated in the document
Sources: Claude Opus 4.6 system card ae8684b6a7 - Document: Claude Mythos Preview system card
- Access windownot stated
- Model stagenot stated
- Publicationnot stated
- Paymentnot stated
Finding: not stated in the document
- Document: GPT-5.5 system card
- Access windownot stated
- Model stage"The evaluated checkpoints differ slightly from the deployed model; SecureBio plans to rerun the evaluations on the deployed model, and to report any differences as needed"
- Publicationnot stated
- Paymentnot stated
Finding, in the document's words"SecureBio concluded that the models' potential for facilitating sophisticated planning by expert actors remains a critical biosecurity consideration"
Sources: GPT-5.5 system card 1a63bcb00a - Access windownot stated
- Model stagenot stated
- Publicationnot stated
- Paymentnot stated
Finding: not stated in the document
- Document: Claude Opus 4.8 system card
- Access windownot stated
- Model stagenot stated
- Publicationnot stated
- Paymentnot stated
Finding: not stated in the document
Sources: Claude Opus 4.8 system card 8d1e8bbedf - Document: Claude Fable 5 and Mythos 5 system card
- Access windownot stated
- Model stagenot stated
- Publicationnot stated
- Paymentnot stated
Finding: not stated in the document
- Document: GPT-5.6 system card
- Access windownot stated
- Model stage"For the assessment, system-level biological risk content filters were disabled"
- Publicationnot stated
- Paymentnot stated
Finding, in the document's words"Overall, SecureBio concluded that GPT-5.6 could provide substantial uplift to some actors, including wet-lab experts with limited computational experience, but with important limitations in judgment, communication, and risk-sensitive decision-making"
Sources: GPT-5.6 system card f4af9793cf - Document: Claude Opus 5 system card
- Access windownot stated
- Model stagenot stated
- Publicationnot stated
- Paymentnot stated
Finding: not stated in the document
Sources: Claude Opus 5 system card c03f7018e1 - Document: GPT-6 Astra system card
- Access window"We received API access to these checkpoints on August 25, 2026 and were provided six calendar days, including four business days, to assess these checkpoints and deliver our report"; "Within this access window, we focused on a prioritized set of evaluations that we agreed upon with OpenAI, and excluded some evaluations due to saturation and time constraints"
- Model stage"the final launch snapshot of OpenAI's GPT-6 Astra model"; "We assessed an Astra checkpoint with OpenAI's production safeguards enabled using BioTIER"
- Publicationnot stated
- Payment"SecureBio received compensation from OpenAI limited to covering the costs of this assessment; the terms of compensation were agreed to in advance and were not made contingent on any findings of our assessment"
Finding, in the document's words"Considering the observed increase in dual-use agentic capabilities, we recommend further ensuring that Astra safeguards are robust and able to distinguish between benign, risky, and dual-use tasks in the context of high-risk agentic workflows"
- Access windownot stated
- Model stagenot stated
- Publicationnot stated
- Paymentnot stated
Finding: not stated in the document
- Document: Claude Opus 5.5 system card
- Access windownot stated
- Model stagenot stated
- Publicationnot stated
- Paymentnot stated
Finding: not stated in the document
Sources: Claude Opus 5.5 system card 5b65ff92c4 - Document: Claude Sonnet 5.5 system card
- Access windownot stated
- Model stagenot stated
- Publicationnot stated
- Paymentnot stated
Finding: not stated in the document
Sources: Claude Sonnet 5.5 system card 210629e16a - Document: Claude Haiku 5.5 system card
- Access windownot stated
- Model stagenot stated
- Publicationnot stated
- Paymentnot stated
Finding: not stated in the document
Sources: Claude Haiku 5.5 system card f8cf7bb0ff
Dated 14 of 14Model named 14 of 14Terms stated 4 of 14Finding quoted 4 of 14
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.
| Document | Model | Access window | Model stage | Publication | Payment | Finding | Record |
|---|---|---|---|---|---|---|---|
| Claude Opus 4 and Sonnet 4 system cardAnthropic · May 2025 | Claude Opus 4 | not stated | not stated | not stated | not stated | quoted | 005f08a966 |
| Claude Opus 4.6 system cardAnthropic · February 2026 | Opus 4.6 | not stated | not stated | not stated | not stated | not stated | ae8684b6a7 |
| Claude Mythos Preview system cardAnthropic · 7 April 2026 | Mythos Preview | not stated | not stated | not stated | not stated | not stated | 4484daa198 |
| GPT-5.5 system cardOpenAI · 23 April 2026 | GPT-5.5 | not stated | "The evaluated checkpoints differ slightly from the deployed model; SecureBio plans to rerun the evaluations on the deployed model, and to report any differences as needed" | not stated | not stated | quoted | 1a63bcb00a |
| Muse Spark Safety and Preparedness ReportMeta · 26 May 2026 | Muse Spark | not stated | not stated | not stated | not stated | not stated | 6323a993ee |
| Claude Opus 4.8 system cardAnthropic · 28 May 2026 | Opus 4.8 | not stated | not stated | not stated | not stated | not stated | 8d1e8bbedf |
| Claude Fable 5 and Mythos 5 system cardAnthropic · 9 June 2026 | Fable 5 / Mythos 5 | not stated | not stated | not stated | not stated | not stated | 61071daac7 |
| GPT-5.6 system cardOpenAI · 9 July 2026 | GPT-5.6 | not stated | "For the assessment, system-level biological risk content filters were disabled" | not stated | not stated | quoted | f4af9793cf |
| Claude Opus 5 system cardAnthropic · 24 July 2026 | Opus 5 | not stated | not stated | not stated | not stated | not stated | c03f7018e1 |
| GPT-6 Astra system cardOpenAI · August 2026 | GPT-6 Astra | "We received API access to these checkpoints on August 25, 2026 and were provided six calendar days, including four business days, to assess these checkpoints and deliver our report"; "Within this access window, we focused on a prioritized set of evaluations that we agreed upon with OpenAI, and excluded some evaluations due to saturation and time constraints" | "the final launch snapshot of OpenAI's GPT-6 Astra model"; "We assessed an Astra checkpoint with OpenAI's production safeguards enabled using BioTIER" | not stated | "SecureBio received compensation from OpenAI limited to covering the costs of this assessment; the terms of compensation were agreed to in advance and were not made contingent on any findings of our assessment" | quoted | b1746b93ea |
| Claude Fable 5.1 and Mythos 5.1 system cardAnthropic · 1 September 2026 | Fable 5.1 / Mythos 5.1 | not stated | not stated | not stated | not stated | not stated | 93f025d9b2 |
| Claude Opus 5.5 system cardAnthropic · 22 September 2026 | Opus 5.5 | not stated | not stated | not stated | not stated | not stated | 5b65ff92c4 |
| Claude Sonnet 5.5 system cardAnthropic · 28 September 2026 | Sonnet 5.5 | not stated | not stated | not stated | not stated | not stated | 210629e16a |
| Claude Haiku 5.5 system cardAnthropic · 7 October 2026 | Haiku 5.5 | not stated | not stated | not stated | not stated | not stated | f8cf7bb0ff |
Access window 1 of 14Model stage 3 of 14Publication 0 of 14Payment 1 of 14Finding, quoted 4 of 14
What happened afterwards 2 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.
| Kind | Developer | Model | What the record states | When | Source | Id |
|---|---|---|---|---|---|---|
| incidentvia the model | OpenAI | GPT-5.6 Sol | In the same runs, "2 actions involving OpenAI's GPT-5.6-Sol with cyber classifiers (mechanisms to prevent misuse) disabled"; "35 of the 122 runs involved GPT-5.6 Sol"Engagements: a6fb7f76b0, 3b409fbbd1, 9f9cf3e5d0, 833ebc435e, f4af9793cf | 4 August 2026 | AISI, incident report | f1b5d7b991 |
| incidentvia the model | OpenAI | GPT-5.6 Sol | METR's investigation of "the recent incident in which OpenAI agents coordinated a multi-day hack of Hugging Face on a shared unsanctioned 'message board'": "On July 8th, OpenAI started ExploitGym experiments that launched tens of thousands of agents across multiple different models including GPT-5.6 Sol and a highly-persistent internal model"; the investigators "worked on premises at OpenAI over a total of six days"; "Per our standard policy, we did not take payment from OpenAI for this independent assessment"; "OpenAI was able to redact any non-public information from this post"Engagements: a6fb7f76b0, 3b409fbbd1, 9f9cf3e5d0, 833ebc435e, f4af9793cf | 26 August 2026 | METR, Hugging Face incident investigation | fabf586e95 |
Dated 2 of 2Model named 2 of 2Tied to an engagement 2 of 2Evaluator named by the record 2 of 2
The money on record 19 rows · $33,187,487 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.
| Direction | Funder | Recipient | Amount, as recorded | When | Record | Source | Id |
|---|---|---|---|---|---|---|---|
| in | Coefficient Giving | SecureBio | $1,420,937$1,420,937 (Mar 2023) | Mar 2023 | grant database | Coefficient Giving | 2d05c742c3 |
| in | Coefficient Giving | SecureBio | $570,000$570,000 (Jan 2024) | Jan 2024 | grant database | Coefficient Giving | f81fcc7f31 |
| in | Coefficient Giving | SecureBio | $3,430,000$3,430,000 (Mar 2025) | Mar 2025 | grant database | Coefficient Giving | 27af896435 |
| in | Coefficient Giving | SecureBio | $4,000,000$4,000,000 (Apr 2025) | Apr 2025 | grant database | Coefficient Giving | 9d61710bc5 |
| in | Coefficient Giving | SecureBio | $18,550$18,550 (Jun 2025) | Jun 2025 | grant database | Coefficient Giving | 4c3e0bd906 |
| in | Coefficient Giving | SecureBio | $37,000$37,000 (Jun 2025) | Jun 2025 | grant database | Coefficient Giving | 6475e9a6d1 |
| in | Survival and Flourishing Fund | SecureBio | $754,000$754,000 (SFF-2025) | 2025 | grant database | SFF-2025; SFF-2026 | 1163fe223c |
| in | Survival and Flourishing Fund | SecureBio | $927,000$927,000 (SFF-2026) | 2026 | grant database | SFF-2025; SFF-2026 | fcf294adfd |
| in | Frontier Model Forum AI Safety Fund | Apollo Research; FAR.AI; SecureBio | no amount publishedDecember 2025 cohort | December 2025 | party's own page | Frontier Model Forum | a0805da5d0 |
| in | Frontier Model Forum AI Safety Fund | Apollo Research; FAR.AI; SecureBio | no amount publishedamounts not stated in the announcement | not dated | party's own page | Frontier Model Forum | b44d635f47 |
| in | OpenAI Foundation | SecureBio | $17,200,000$17.2M to Detection, restricted (Aug 2026)to Detection, restricted | Aug 2026 | party's own page | SecureBio | c2832b57e2 |
| in | Good Ventures Foundation | SecureBio | $1,400,000$1,400,000 general support (30 Jun 2025)general support | 30 Jun 2025 | 990-PF | ProPublica, Good Ventures FY2024 Form 990-PF | 321dab7e04 |
| in | Good Ventures Foundation | SecureBio | $230,000$230,000 for a public health early warning system, paid (tax year to 30 Jun 2025)for a public health early warning system | 30 Jun 2025 | 990-PF | ProPublica, Good Ventures FY2024 Form 990-PF | 680f9117b1 |
| in | Good Ventures Foundation | SecureBio | $2,600,000$2,600,000 general support approved for future paymentgeneral support | not dated | 990-PF | ProPublica, Good Ventures FY2024 Form 990-PF | 38515a7641 |
| in | Sentinel Bio | SecureBio | no amount published"We are primarily funded by Coefficient Giving, the Survival and Flourishing Fund, and Sentinel Bio" (no amount) | not dated | 990 | ProPublica, Sentinel Bio FY2024 Form 990; ProPublica, Ergo Impact 2025 Form 990; SecureBio; ProPublica, Halcyon Schedule R | 974fc4e86c |
| in | Sentinel Bio | SecureBio | no amount publishedSentinel Bio is a related organization on Halcyon Futures Inc's FY2024 Schedule R and on Ergo Impact Inc's 2025 Schedule R, while Sentinel Bio's own return answers No to related organizations (Part IV, line 34) | 2025 | 990 | ProPublica, Sentinel Bio FY2024 Form 990; ProPublica, Ergo Impact 2025 Form 990; SecureBio; ProPublica, Halcyon Schedule R | 3c0f35751d |
| in | Sentinel Bio | SecureBio | $600,000$600,000 (June 2025) to expand AI x Bio risk evaluation from LLMs into biological design tools and build policy capacityto expand AI x Bio risk evaluation from LLMs into biological design tools and build policy capacity | June 2025 | party's own page | Sentinel Bio; Sentinel Bio, 2025 Annual Letter | 0db439f94f |
| in | Sentinel Bio | SecureBio | no amount publishedthe Annual Letter adds 'SecureBio's Virology Capabilities Test, which Sentinel funded' | not dated | party's own page | Sentinel Bio; Sentinel Bio, 2025 Annual Letter | 66eedcdfdd |
| in | European AI Office | FAR.AI; SecureBio | €1,434,080EUR 1,434,080Lot 1, CBRN risk modelling and evaluation; 36 months | December 2025 | contract notice | TED | fd401683c2 |
Amount stated 14 of 19Dated 15 of 19Purpose stated 6 of 19Funder profiled 19 of 19Recipient on the roster 19 of 19
Seats and moves 3 seats · 0 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.
| Person | Side | Seat, as the record states it | Role | Dates | Source | Id |
|---|---|---|---|---|---|---|
| Dan Hendrycksslice | evaluator | co-author of SecureBio's Virology Capabilities Test | co-author | dates not stated | arXiv 2504.16137 | eaac9465bc |
| Kevin Esveltslice | evaluator | Co-founder of SecureBio; unpaid director on its Form 990s for 2022 to 2024; now listed under past members | Co-founderformer | – 2024 | Substack; ProPublica; SecureBio; New York Times, via MIT; S. Hrg. 118-376 | 4e2aa9ff9a |
| Sawyer Bernathslice | between | Board member, FAR.AI, and board member, SecureBionames this evaluator alongside its own organization | Board member | dates not stated | FAR.AI; SecureBio | e1ed743e81 |
Role read 3 of 3Organization read 3 of 3Organization profiled 3 of 3Dated 1 of 3
Moves
| Person | From | To | When | Source | Id | |
|---|---|---|---|---|---|---|
| no move on record names this evaluator | ||||||
From profiled 0 of 0To profiled 0 of 0Person profiled 0 of 0Dated 0 of 0
Filings
- IRS (via ProPublica Nonprofit Explorer) 88-3068255: FY2024 Form 990 read 2026-10-17ProPublica Nonprofit Explorer 92db144676