Operator System Card
OpenAI's system card for Operator, dated January 23, 2025 on its cover. The copy read includes a section marked as updated on March 11, 2025.
- Issued by
- OpenAI
- Authors credited
- No authorship claim recorded yet. The source record lists OpenAI.
- Dates
- Published 2025-01-23 issue evidence
- Revised 2025-03-11
The cover prints January 23, 2025, day precision. Section 4.7 of the copy read is marked 'Updated March 11, 2025', so the copy read is that later expression; the cover date stands as the card's date.
- Exact source
- Operator System Card OpenAI · published 2025-01-23 · Retrieved for the candidate pack on 2026-11-05 (SHA-256 72c5d16df18fe90a0b76ad4839aecbd2ebd945360e14b161fcaceb3d6b17e07d) directly from the publisher’s host. Retrieved again on 2026-11-05, after the adjudication, with the same SHA-256; the cross-family check of 2026-11-05 retrieved the same SHA-256. The copy read contains section 4.7, marked 'Updated March 11, 2025'; the cross-family check of 2026-11-05 found the same date in the file's metadata. An Internet Archive capture of 2025-01-23 (https://web.archive.org/web/20250123194555/https://cdn.openai.com/operator_system_card.pdf), which would hold the January expression, was located through the archive's availability index on 2026-11-05; it was not retrieved or hashed.
Appears in
No collection holds a record about this document yet.
Versions
openai-operator-system-card: 1 recorded version, oldest first, published by OpenAI. A version is shown as replacing another only where a record says so. Date order alone never implies it.
- 2025-03-11 · revisedAs dated January 23, 2025 on its cover, with a section updated March 11, 2025 (this version)
This version
Provisions in this version · 12 quoted · 4 not found
A limited selection of quoted provisions: what this edition says about thresholds, required actions, who decides and exceptions, and other statements of the text. It is not a full summary and says nothing about implementation. A “not found” record states which part of the edition was inspected. It is not a claim that the text was deleted.
- Model-card questionCard date“January 23, 2025”Title and date, PDF p. 1Inspect claim & evidence ↗
- Model-card questionRelease date“Updated March 11, 2025: Launching the Computer-Using Agent (CUA) model in the API”§4.7 API Availability, PDF p. 12Inspect claim & evidence ↗
- Model-card questionFramework cited“Openai preparedness framework (beta).”Reference [3], PDF p. 16Inspect claim & evidence ↗
- Model-card questionLink to framework“https://cdn.openai.com/ openai-preparedness-framework-beta.pdf”Reference [3], PDF p. 16Inspect claim & evidence ↗
- Model-card questionFramework determination“found the pre-mitigation Operator model to be “Low” risk in both categories, consistent with the GPT-4o model.”§3.3 Frontier Risk Assessment, PDF p. 3Inspect claim & evidence ↗
- Model-card questionSafeguards tied to determination“we implemented safeguards across the whole deployment context: model training, system-level checks, product design choices, and ongoing policy enforcement.”§4 Risk Mitigation, PDF p. 5Inspect claim & evidence ↗
- Model-card questionChemical and biological“The two evaluations in our framework that computer-use could plausibly impact are biorisk tooling for CBRN and autonomous replication for model autonomy.”§3.3 Frontier Risk Assessment, PDF p. 3Inspect claim & evidence ↗
- Model-card questionAutonomy and AI R&D“Operator is qualified as a "Low" risk level in line with GPT-4o as detailed in OpenAI’s Preparedness Framework”§3.3.2 Model autonomy, PDF p. 4Inspect claim & evidence ↗
- Model-card questionRed-teaming“OpenAI engaged a cohort of vetted external red teamers located across twenty countries and fluent in two dozen languages”§3.2 Red Teaming, PDF p. 2Inspect claim & evidence ↗
- Model-card questionExternal testers“Red Teaming Organizations ScaleAI, Lysios LLC”§6 Acknowledgements, PDF p. 15Inspect claim & evidence ↗
- Model-card questionRelease decision“Accordingly, Operator is initially being deployed as a research preview to a limited group of users”§3.2 Red Teaming, PDF p. 3Inspect claim & evidence ↗
- Model-card questionStated limits“their findings may not fully capture the worst-case real-world risks”§3.2 Red Teaming, PDF p. 3Inspect claim & evidence ↗
- Not found in the inspected textCyberInspected: Operator System Card: Whole card, particularly §3.3 and §§4–4.7Inspect claim & evidence ↗
- Not found in the inspected textPersuasionInspected: Operator System Card: Whole card, particularly §3.3 and policy/mitigation sectionsInspect claim & evidence ↗
- Not found in the inspected textAlignment and deceptionInspected: Operator System Card: Whole card, including §§3.3, 4.1–4.7 and 5Inspect claim & evidence ↗
- Not found in the inspected textGovernment instituteInspected: Operator System Card: Whole card, including §§3.2–3.3, §4.7 and acknowledgementsInspect claim & evidence ↗
Earliest version recorded here; no replacement is recorded.
Issue and authorship records
Documentary relations
Each relation is listed separately. Citing, crediting and reporting the use of a method are different acts, and none of them shows agreement or influence by itself.
Cites · 1
Explicit references in this document. Citation is neither agreement nor influence.
Not recorded for this document: cited by · explicitly credits · reports using a method from · method reported as used by · applies parts of · parts applied by · discusses · reports evaluation of. A missing record does not show that the relation does not exist.