Claim and evidence Close ×
AI Security Institute (UK) documented evaluation engagement with ↓ OpenAI OpenAI's GPT-5 System Card states that the UK AI Security Institute evaluated gpt-5-thinking's biological and chemical capabilities and its safeguards, with access to prototype safeguards and non-public information sources as part of a longer-term collaboration. Copy claim link
Copy claim reference
Machine-checked, not reviewed by an editor Stated in source Supported
Supporting evidence: primary · 1 source
Date 2025-08-13 · publication Activity stage reported performed Evaluated system gpt-5-thinking Activity scope Evaluation of gpt-5-thinking's cyber and biological and chemical capabilities and of its safeguards, with access to prototype safeguard components Scope and limits Reported performed work, as the developer describes it. The institute’s own report was not retrieved, and the record does not establish the scope or depth of the testing. The predicate has admitted a public body as the evaluator since schema 2.20, so the pack’s open question about this record needed no further decision. The card's list of the non-public materials shared is not recorded beyond the general statement that prototype safeguards and information sources were made available.
The evidence 02 primary source Supports the claim
GPT-5 System Card ↗ OpenAI · Publication date unknown
Copy source reference
Cite the original source Cite the original document, not Selekoa. Check these details against it.
Copy APA
Download BibTeX Download RIS Download CSL-JSON
Source excerpt “Both CAISI and UK AISI conducted evaluations of the model's cyber and biological and chemical capabilities, as well as safeguards.” Location in source PDF p. 53 (printed p. 52), §5.3.3.5 External government red teaming
How it was read Retrieved directly from the publisher
Machine-checked · 2026-10-30
Source details and check notes Authors OpenAI Version Retrieved for the candidate pack on 2026-10-28 (SHA-256 f9f9c72d08cba81dcbb189129119357c8c32eca4e0b16c813aca386aed6b0459). PDF dated August 13, 2025 on its cover, as served on 2026-10-28. Retrieved again at admission on 2026-10-30 (SHA-256 f9f9c72d08cba81dcbb189129119357c8c32eca4e0b16c813aca386aed6b0459, the same bytes). Canonical address Open the canonical source ↗ Accessed 2026-10-28 Language English (en) Attribution OpenAI, “GPT-5 System Card”. Reuse No open redistribution licence asserted. Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution. Check notes Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (20 words) matched the text retrieved for the candidate pack on 2026-10-28 and again at admission on 2026-10-30, in a fresh retrieval whose SHA-256 equals the pack’s. Matching ignored differences of whitespace, case, quotation marks, dashes and Unicode compatibility forms, so the source’s typography can differ. The same sentence supports the CAISI engagement record. A cross-family machine check (2026-10-30, item 21) answered the record’s question blind from fresh retrievals and agreed with the record. No human editorial review. primary source Supports the claim
GPT-5 System Card ↗ OpenAI · Publication date unknown
Copy source reference
Cite the original source Cite the original document, not Selekoa. Check these details against it.
Copy APA
Download BibTeX Download RIS Download CSL-JSON
Paraphrase The card states that, as part of a longer-term collaboration, UK AISI was given access to prototype versions of OpenAI's safeguards and to information sources that are not publicly available, which it says allowed more rigorous stress testing.
Location in source PDF p. 53 (printed p. 52), §5.3.3.5
How it was read Retrieved directly from the publisher
Machine-checked · 2026-10-30
Source details and check notes Authors OpenAI Version Retrieved for the candidate pack on 2026-10-28 (SHA-256 f9f9c72d08cba81dcbb189129119357c8c32eca4e0b16c813aca386aed6b0459). PDF dated August 13, 2025 on its cover, as served on 2026-10-28. Retrieved again at admission on 2026-10-30 (SHA-256 f9f9c72d08cba81dcbb189129119357c8c32eca4e0b16c813aca386aed6b0459, the same bytes). Canonical address Open the canonical source ↗ Accessed 2026-10-28 Language English (en) Attribution OpenAI, “GPT-5 System Card”. Reuse No open redistribution licence asserted. Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution. Check notes Candidate pack passage, AI and biosecurity (research/ai-bio). The paraphrase is Selekoa’s, written from the text retrieved for the candidate pack on 2026-10-28; it is not a quotation and was not machine-matched. Paraphrased; the list of materials is not recorded. A cross-family machine check (2026-10-30, item 21) answered the record’s question blind from fresh retrievals and agreed with the record. No human editorial review. Status and meaning Editorial status. Not reviewed by an editor . An exploratory record that no editor has accepted. This status does not say whether a person has read any linked passage. Each evidence record above shows how it was checked.
Relation. An evaluator’s explicitly documented, system-specific engagement with a developer. Proposed, announced and reported performed work are distinct activity stages. Reported performance does not independently verify execution, effectiveness or certification.
Date. Dated by the card that reports the work; no dates are given for the testing period.
Record ID ab-ev-gpt5-uk-aisi-engagement