OpenAI · GPT-4o System Card (PDF)

CBRN Frontier Risk Evaluation

High severity
Share 𝕏 Share in Share 🔒 PDF

What it is

OpenAI tested whether GPT-4o could help someone create weapons of mass destruction and concluded the risk is 'medium' — serious enough to track closely but not yet at the level that would have blocked the model's release.

Consumer impact (what this means for users)

While this risk primarily affects society at large rather than individual users, it signals that GPT-4o has capabilities that could be misused for serious harm, and OpenAI's internal threshold — not a regulatory one — is the only current gate on deployment.

Cross-platform context

See how other platforms handle CBRN Frontier Risk Evaluation and similar clauses.

Compare across platforms →
Need full compliance memos? See Professional →

Why it matters (compliance & risk perspective)

A 'medium' CBRN risk rating under OpenAI's own framework means GPT-4o can provide meaningful assistance to individuals attempting to create weapons capable of mass casualties — this is a significant public safety disclosure.

View original clause language
We evaluated GPT-4o for its potential to provide serious uplift to those seeking to create biological, chemical, nuclear, or radiological weapons with the potential for mass casualties. We assessed GPT-4o to be Medium risk per our Preparedness Framework... While the model has the potential to meaningfully uplift those seeking to create biological, chemical, nuclear, or radiological weapons, the level of uplift provided does not yet meet our threshold for Critical risk.

Institutional analysis (Compliance & legal intelligence)

(1) REGULATORY FRAMEWORK: This provision implicates US Executive Order 14110 on Safe, Secure, and Trustworthy AI (dual-use foundation model reporting requirements), the EU AI Act Article 5 (prohibited AI practices that risk public safety), and potentially the Biological Weapons Anti-Terrorism Act and Chemical Weapons Convention domestic implementing statutes. The Department of Homeland Security and the National Security Council have oversight interests. The EU AI Office has enforcement authority for prohibited practices under the EU AI Act. (2)

🔒

Compliance intelligence locked

Regulatory citations, enforcement risk, and due diligence action items.

Watcher $9.99/mo Professional $149/mo

Watcher: regulatory citations. Professional: full compliance memo.

Applicable agencies

  • FTC
    FTC consumer protection authority is relevant where CBRN capability disclosures are insufficiently prominent in consumer-facing materials, potentially constituting deceptive omission.
    File a complaint →

Provision details

Document information
Document
GPT-4o System Card (PDF)
Entity
OpenAI
Document last updated
March 5, 2026
Tracking information
First tracked
March 10, 2026
Last verified
April 27, 2026
Record ID
CA-P-003141
Document ID
CA-D-00008
Evidence Provenance
Source URL
Wayback Machine
SHA-256
7c23ef53467eea199596abe78511d57ffee1e94b50ef10ac0f7d81df278b5059
Verified
✓ Snapshot stored   ✓ Change verified
How to Cite
ConductAtlas Policy Archive
Entity: OpenAI | Document: GPT-4o System Card (PDF) | Record: CA-P-003141
Captured: 2026-03-10 03:40:55 UTC | SHA-256: 7c23ef53467eea19…
URL: https://conductatlas.com/platform/openai/gpt-4o-system-card-pdf/cbrn-frontier-risk-evaluation/
Accessed: May 2, 2026
Classification
Severity
High
Categories

Other provisions in this document