Get the weekly 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. No account.
This provision authorizes Duolingo to record and store audio and text submitted through AI-powered features including Video Call, and to use those recordings and transcripts for product improvement and AI model training. The text and audio may also be shared with vendors including OpenAI and Google, subject to contractual restrictions on those vendors' independent use of the data.
This analysis describes what Duolingo'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 authorizes Duolingo to retain AI interaction content and use it for AI model training purposes, and to share that content with named third-party AI vendors. The adequacy of the vendor contractual restrictions on secondary use may warrant independent compliance review, particularly under GDPR data minimization and purpose limitation principles.
The updated policy removes explicit language stating that Android users and website users are not subject to audio collection for product improvement purposes. Previously, the policy authorized audio collection only from iOS users, with an explicit carve-out for Android and web users. The revised language now states that all users may choose not to share audio within app Settings, suggesting audio collection may now occur across all platforms unless the opt-out mechanism is used. The practical operational effect of this change depends on whether Duolingo implements audio collection on Android and web platforms, which the policy change does not explicitly confirm. You can decline audio sharing for product improvement by adjusting the setting within the app.
View change record →Under this provision, audio and text submitted during Video Call and other AI features may be shared with OpenAI and Google and stored by Duolingo for product improvement and AI model training. The policy advises users not to submit personal, sensitive, or confidential information in these interactions.
Cross-platform context
See how other platforms handle AI Audio Recording and Model Training and similar clauses.
Compare across platforms →Monitoring
Duolingo has changed this document before.
Receive same-day alerts, structured change summaries, and monitoring for up to 25 platforms.
"When you interact with Video Call or other AI-enabled features, the text and audio you submit may be shared with AI vendors such as OpenAI and Google. Duolingo's agreements with these AI vendors mean they are not permitted to use any personal information for their own purposes. Additionally, Duolingo may generate, record, and store audio recordings or transcripts of the text and audio you submit, and use these recordings or transcripts for product improvement and personalization purposes, including training and running Duolingo's own artificial intelligence models. Please do not submit any personal, sensitive, or confidential information when using Video Call or other AI features.Excerpt from Duolingo's Privacy Policy
1) REGULATORY LANDSCAPE: This provision implicates GDPR purpose limitation and data minimization principles, particularly regarding the retention and secondary use of audio recordings for AI training. The EU AI Act may impose additional obligations on AI system training data practices. The FTC Act's unfair or deceptive practices framework is relevant to the accuracy of vendor restriction claims. COPPA is relevant because the policy separately states that Child Users' speech data is not shared with Duolingo for product improvement, indicating this provision applies to adult users. 2) GOVERNANCE EXPOSURE: Medium-High. The provision asserts that vendor agreements restrict OpenAI and Google from using personal information for their own purposes, but the policy does not disclose the specific contractual terms or provide audit rights. If vendor processing exceeds the stated restrictions, Duolingo's DPF liability framework would apply. The policy's self-advisory instruction to not submit sensitive information does not constitute a technical control. 3) JURISDICTION FLAGS: EU and EEA users face the most significant exposure under GDPR Article 5 data minimization and Article 13/14 transparency requirements regarding AI training use. UK users are subject to equivalent UK GDPR obligations. California users may have CCPA rights relevant to AI-processed personal data. Illinois users should note that if audio processing constitutes biometric data capture, the Illinois Biometric Information Privacy Act may apply, though this depends on how speech data is processed. 4) CONTRACT AND VENDOR IMPLICATIONS: Procurement teams should request copies of Duolingo's data processing agreements with OpenAI and Google to verify the scope of the restrictions claimed in this policy. The DPF liability clause in the policy states that Duolingo remains liable if its agents process data inconsistently with DPF Principles, which provides a contractual basis for liability but requires affirmative enforcement by affected individuals. 5) COMPLIANCE CONSIDERATIONS: Legal teams should verify whether consent mechanisms for AI feature use satisfy GDPR Article 7 requirements for freely given, specific, informed consent where AI training is a stated purpose. Data mapping should reflect audio and transcript flows to OpenAI and Google. Retention schedules for AI-generated transcripts and recordings should be documented and reviewed against data minimization obligations.
This provision authorizes Duolingo to retain AI interaction content and use it for AI model training purposes, and to share that content with named third-party AI vendors. The adequacy of the vendor contractual restrictions on secondary use may warrant independent compliance review, particularly under GDPR data minimization and purpose limitation principles.
Under this provision, audio and text submitted during Video Call and other AI features may be shared with OpenAI and Google and stored by Duolingo for product improvement and AI model training. The policy advises users not to submit personal, sensitive, or confidential information in these interactions.
No. ConductAtlas is an independent monitoring service. We are not affiliated with, endorsed by, or sponsored by Duolingo.