OpenAI · GPT-4o System Card (PDF)

Incomplete Audio Modality Mitigations at Launch

High severity
Share 𝕏 Share in Share 🔒 PDF

What it is

OpenAI launched GPT-4o's voice features knowing that safety measures for audio were not yet fully complete, and intentionally limited access while they continue developing those protections.

Consumer impact (what this means for users)

Users who interact with GPT-4o through voice interfaces may be exposed to risks — including voice impersonation and emotionally manipulative outputs — that OpenAI has not yet fully mitigated, and the level of protection depends on which platform or operator they use.

Cross-platform context

See how other platforms handle Incomplete Audio Modality Mitigations at Launch and similar clauses.

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

Why it matters (compliance & risk perspective)

This is a rare public admission by a major AI company that a consumer-facing feature was released with known, unresolved safety gaps — meaning users of voice features have less protection than users of text features.

View original clause language
GPT-4o's audio modality introduces new risks that require new mitigations. We are still developing and testing our audio safety mitigations and as a result, we are launching with a limited release of audio outputs, including voice... We are continuing to work on audio safety research and will expand access to audio over time as we develop and validate additional mitigations.

Institutional analysis (Compliance & legal intelligence)

(1) REGULATORY FRAMEWORK: This provision implicates FTC Act Section 5 (unfair or deceptive practices — deploying a product with known unmitigated harm vectors), EU AI Act Article 9 (risk management system obligations requiring iterative risk assessment throughout lifecycle), and EU AI Act Article 13 (transparency obligations requiring disclosure of known limitations). The FTC has enforcement authority in the US; the European AI Office oversees EU AI Act compliance. (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 Section 5 authority covers unfair or deceptive practices, including deploying AI products with known unmitigated consumer harm vectors without adequate point-of-use disclosure.
    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-003140
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-003140
Captured: 2026-03-10 03:40:55 UTC | SHA-256: 7c23ef53467eea19…
URL: https://conductatlas.com/platform/openai/gpt-4o-system-card-pdf/incomplete-audio-modality-mitigations-at-launch/
Accessed: May 2, 2026
Classification
Severity
High
Categories

Other provisions in this document