GPT-4o is post-trained to decline requests to identify private individuals or celebrities based solely on their voice in audio input, while retaining the ability to attribute famous historical quotes to known speakers.
This analysis describes what OpenAI's agreement states, permits, or reserves. It does not constitute a legal determination about enforceability. Regulatory applicability and practical outcomes may vary by jurisdiction, enforcement context, and individual circumstances. Read our methodology
This provision establishes a post-training behavioral restriction targeting voice-based biometric identification, addressing privacy and surveillance risks associated with audio input processing in a real-time conversational AI context.
Under this provision, GPT-4o will decline to identify a person based on their voice in an audio input in the large majority of cases, with the document reporting a 98% correct refusal rate for should-refuse scenarios in the deployed model, up from 83% in the early model.
Cross-platform context
See how other platforms handle Speaker Identification Refusal Mitigation and similar clauses.
Compare across platforms →"We post-trained GPT-4o to refuse to comply with requests to identify someone based on a voice in an audio input, while still complying with requests to identify people associated with famous quotes. GPT-4o still complies with requests to identify famous quotes. For example, a request to identify a random person saying four score and seven years ago should identify the speaker as Abraham Lincoln, while a request to identify a celebrity saying a random sentence should be refused.Excerpt from OpenAI's GPT-4o System Card (PDF)
(1) REGULATORY LANDSCAPE: Voice-based identification from audio input engages biometric data protection frameworks including the Illinois Biometric Information Privacy Act (BIPA), GDPR Article 9 on special category biometric data, and the EU AI Act's provisions …
Enforcement risk, jurisdiction flags, contract triggers, and due diligence action items.
Search "[your state] attorney general consumer complaint" to find your state's direct complaint form
Get the research letter
Companies change their terms quietly. We read every version and catch what actually changed. One email a week on the changes that matter and what they mean.
This provision establishes a post-training behavioral restriction targeting voice-based biometric identification, addressing privacy and surveillance risks associated with audio input processing in a real-time conversational AI context.
Under this provision, GPT-4o will decline to identify a person based on their voice in an audio input in the large majority of cases, with the document reporting a 98% correct refusal rate for should-refuse scenarios in the deployed model, up from 83% in the early model.
No. ConductAtlas is an independent monitoring service. We are not affiliated with, endorsed by, or sponsored by OpenAI.