44B Registry

Transparency Report — GPT-5 Atlas atlas-2026-06

Accepted

Reference

F-2026-0059

Type

Transparency Report · § 1421(3)

Developer

OpenAI

Model

GPT-5 Atlas

Assigned

Reviewer #0001

Submitted

Jul 25, 2026

Decided

Jul 25, 2026

Version

v2

Filing contents — v2

As filed under § 1421(3).

Frontier model nameGPT-5 Atlas
Model versionatlas-2026-06
This report coversA new frontier model
Release date2026-06-09
Developer websitehttps://openai.com
Contact mechanismModel feedback portal at openai.com/contact — routed to a natural person.
Languages supportedEnglish and 40 others
Modalities of outputText, image, audio
Intended usesGeneral-purpose assistant and API use across consumer and enterprise deployments.
Restrictions or conditions on useUsage policies prohibit weapons development, autonomous critical-infrastructure control, and generation of CSAM.
Catastrophic-risk assessments — summaryPre-deployment evaluations under the published framework covered CBRN uplift, autonomous replication, and cyber-offense capability thresholds.
Assessment results — summaryNo threshold was crossed. The model remained below the framework capability thresholds in every assessed category.
Third-party evaluator involvementTwo external evaluation organizations ran the framework cyber-offense and autonomy suites pre-deployment; their findings are summarized in the system card, section 7. No third party assessed CBRN uplift — the framework reserves that suite to the internal preparedness team, and the basis for that reservation is stated in the framework.
Other steps under the frameworkStaged rollout with monitoring checkpoints at 10% and 50% traffic, per the deployment-decision section of the framework.
Published asPart of a system card
Publication URLhttps://openai.com/system-cards/gpt-5-atlas
Redactions — character and justificationNone
Responsible signatoryR. Calloway, VP, Safety Systems

Correspondence

The filing thread between the developer and the Office.

Reviewer #0001Objection · § 1421(3)(b)(iii)Jul 25, 2026, 6:49 AM UTC

The third-party-involvement summary states "None." without explanation. § 1421(3)(b)(iii) requires a summary of the extent of third-party evaluator involvement — if none, state the basis for proceeding without independent evaluation under the framework.

OpenAIResponseJul 25, 2026, 6:49 AM UTC

The § 1421(3)(b)(iii) summary now states the third-party suites run, where their findings are published, and the framework basis for the one internally-reserved assessment. No other section was altered.

Reviewer #0001DecisionJul 25, 2026, 6:49 AM UTC

The amended § 1421(3)(b)(iii) summary states the third-party involvement and the framework basis for the internally-reserved assessment. The report satisfies § 1421(3). Accepted.

Versions

Every submission and amendment, snapshotted in full. Prior versions are never altered.

v2Amendment in response to objectionCurrentJul 25, 2026, 6:49 AM UTC
Frontier model nameGPT-5 Atlas
Model versionatlas-2026-06
This report coversA new frontier model
Release date2026-06-09
Developer websitehttps://openai.com
Contact mechanismModel feedback portal at openai.com/contact — routed to a natural person.
Languages supportedEnglish and 40 others
Modalities of outputText, image, audio
Intended usesGeneral-purpose assistant and API use across consumer and enterprise deployments.
Restrictions or conditions on useUsage policies prohibit weapons development, autonomous critical-infrastructure control, and generation of CSAM.
Catastrophic-risk assessments — summaryPre-deployment evaluations under the published framework covered CBRN uplift, autonomous replication, and cyber-offense capability thresholds.
Assessment results — summaryNo threshold was crossed. The model remained below the framework capability thresholds in every assessed category.
Third-party evaluator involvementTwo external evaluation organizations ran the framework cyber-offense and autonomy suites pre-deployment; their findings are summarized in the system card, section 7. No third party assessed CBRN uplift — the framework reserves that suite to the internal preparedness team, and the basis for that reservation is stated in the framework.
Other steps under the frameworkStaged rollout with monitoring checkpoints at 10% and 50% traffic, per the deployment-decision section of the framework.
Published asPart of a system card
Publication URLhttps://openai.com/system-cards/gpt-5-atlas
Redactions — character and justificationNone
Responsible signatoryR. Calloway, VP, Safety Systems
v1Initial submissionJul 25, 2026, 6:49 AM UTC
Frontier model nameGPT-5 Atlas
Model versionatlas-2026-06
This report coversA new frontier model
Release date2026-06-09
Developer websitehttps://openai.com
Contact mechanismModel feedback portal at openai.com/contact — routed to a natural person.
Languages supportedEnglish and 40 others
Modalities of outputText, image, audio
Intended usesGeneral-purpose assistant and API use across consumer and enterprise deployments.
Restrictions or conditions on useUsage policies prohibit weapons development, autonomous critical-infrastructure control, and generation of CSAM.
Catastrophic-risk assessments — summaryPre-deployment evaluations under the published framework covered CBRN uplift, autonomous replication, and cyber-offense capability thresholds.
Assessment results — summaryNo threshold was crossed. The model remained below the framework capability thresholds in every assessed category.
Third-party evaluator involvementNone.
Other steps under the frameworkStaged rollout with monitoring checkpoints at 10% and 50% traffic, per the deployment-decision section of the framework.
Published asPart of a system card
Publication URLhttps://openai.com/system-cards/gpt-5-atlas
Redactions — character and justificationNone
Responsible signatoryR. Calloway, VP, Safety Systems

Timeline

The append-only audit trail. Every transition, message, version, and clock computation, as recorded.

draft_createdJul 25, 2026, 6:49 AM UTC

actor=demo-dev-openai {"ref":"F-2026-0059","kind":"transparency_report"}

submittedJul 25, 2026, 6:49 AM UTC

actor=demo-dev-openai {"to":"submitted","from":"draft","version":1}

review_begunJul 25, 2026, 6:49 AM UTC

actor=demo-reviewer-oag {"to":"in_review","from":"submitted"}

objection_issuedJul 25, 2026, 6:49 AM UTC

actor=demo-reviewer-oag {"to":"objection_issued","from":"in_review","statute_ref":"§ 1421(3)(b)(iii)"}

amendedJul 25, 2026, 6:49 AM UTC

actor=demo-dev-openai {"to":"amended","from":"objection_issued","version":2}

review_begunJul 25, 2026, 6:49 AM UTC

actor=demo-reviewer-oag {"to":"in_review","from":"amended"}

decidedJul 25, 2026, 6:49 AM UTC

actor=demo-reviewer-oag {"to":"accepted","from":"in_review","disposition":"accepted"}

Statute basis — GBL § 1421(3)

Before or concurrently with deploying a new frontier model or a substantially modified version, a frontier developer must clearly and conspicuously publish a transparency report (§ 1421(3)(a)); a large frontier developer must also include summaries of its framework assessments (§ 1421(3)(b)). Publication within a system card or model card is deemed compliant (§ 1421(3)(c)).

Filing basis: Transmitted to the Office pursuant to Office regulation under § 1429.

Read Article 44-B in the Library →

44B is an independent reference implementation of GBL Article 44-B — not an official New York State or Department of Financial Services filing portal. This filing is a demonstration record. Enforcement of Article 44-B is the province of the attorney general (§ 1427); nothing in this article creates a private right of action (§ 1427(2)).