openai.com — 68/100 (Medium privacy risk)
Last analyzed
OpenAI · openai.com
Report details
Medium privacy riskOpenAI’s EU privacy policy offers strong user controls and transparency but falls short on data minimization, third-party sharing clarity, and international transfer safeguards for EU users.
OpenAI’s EU privacy policy provides detailed disclosures about data collection, processing purposes, and user rights, including explicit opt-outs for model training. However, it collects broad categories of data (e.g., user content, logs, device info), shares data extensively with vendors and affiliates, and transfers data to the US and other non-EEA countries using SCCs. While it offers granular controls (e.g., temporary chats, model training opt-outs), gaps remain in data minimization, third-party subprocessor transparency, and the practical enforceability of transfer mechanisms under EU law. The policy is compliant on paper but raises concerns about proportionality and oversight.
Category Assessment
Breakdown of the policy across key compliance areas. Good = strong, fair = mixed, poor = concerning.
Collects extensive data (e.g., prompts, files, IP addresses, device IDs) beyond what’s strictly necessary for core functionality, though some controls (e.g., temporary chats) limit retention.
Detailed disclosures on data types, purposes, and legal bases, with clear links to help center articles and opt-out mechanisms.
Shares data broadly with vendors, affiliates, and business partners (e.g., cloud providers, payment processors) without a comprehensive, named subprocessor list or granular consent for each category.
Uses SCCs for non-EEA transfers but lacks detail on supplementary measures to address US surveillance risks, and does not restrict transfers to ‘adequate’ jurisdictions only.
Explicit opt-out for model training is available, and users can disable memory/history features to limit data used for improvements.
Clearly lists GDPR rights (access, deletion, portability, objection) and provides multiple channels (account settings, privacy portal, email) to exercise them.
Key Findings
Notable clauses, issues, or positive practices discovered (critical first)
Inadequate safeguards for international transfers to the US
Section 10 confirms transfers to the US and other non-EEA countries using SCCs, but does not describe supplementary technical/organizational measures (e.g., end-to-end encryption, zero-knowledge architectures) to address US government surveillance risks. Post-Schrems II, SCCs alone are often deemed insufficient by EU courts and regulators (e.g., CNIL, EDPB).
Overly broad data collection undermines minimization
Section 1 lists extensive categories of data collected, including user content (prompts, files, images, voice/video), log data (IP, browser settings), usage data (interactions, feedback), device info, and location data. While some data is necessary for service delivery, the policy does not justify the proportionality of collecting all these categories for every user or feature. For example, location data is inferred from IP addresses ‘for security’ but also used for ‘product experience’ without clear necessity.
Lack of named subprocessor list for third-party sharing
Section 3 states data is shared with ‘vendors and service providers’ (e.g., hosting, cloud, analytics, payment processors) but does not provide a named list or link to a subprocessor registry. This omits critical transparency required under GDPR Art. 28(3) for users to assess risks of sharing with specific third parties (e.g., AWS, Stripe).
Ambiguous legal basis for model training with user content
Section 2 states user content may be used to ‘improve models’ but relies on ‘legitimate interests’ as a legal basis, which may conflict with GDPR’s requirement for explicit consent for high-risk processing (e.g., training generative AI). While an opt-out exists, the default inclusion of user content in training without clear consent is a compliance risk.
Retention periods lack specificity for certain data types
Section 4 describes retention based on ‘legitimate business purposes’ (e.g., fraud prevention, legal obligations) but does not specify maximum durations for categories like log data or usage analytics. For example, it states data may be retained ‘as long as necessary’ for security but does not define what ‘necessary’ entails for non-critical data.
Consumer Takeaway
You have significant control over your data (e.g., opting out of model training, deleting chats), but OpenAI collects a wide range of personal data, shares it with many third parties, and transfers it outside the EU. If privacy is a priority, disable model training, avoid uploading sensitive files, and review third-party integrations carefully.
Compliance Posture
The policy aligns with GDPR requirements on transparency, legal bases, and user rights, but its reliance on SCCs for non-EEA transfers and broad data sharing with vendors/affiliates may not satisfy stricter interpretations of EU data protection (e.g., Schrems II). The Irish DPC’s oversight adds credibility, but enforcement risks persist for high-risk transfers.
EU Transfers
Data is transferred to the US and other non-EEA countries using Standard Contractual Clauses (SCCs) and, where applicable, adequacy decisions. However, the policy does not address supplementary measures (e.g., encryption, pseudonymization) to mitigate US surveillance risks under FISA 702, leaving transfers vulnerable to legal challenges under EU law. No explicit mention of Binding Corporate Rules (BCRs) or other safeguards beyond SCCs.
Detected Signals
Specific data points and practices identified in the text
Evidence Snippets
Direct quotes from the policy supporting these findings
Vi indsamler personoplysninger om dig (“Personoplysninger”) som følger: Kontooplysninger, Brugerindhold, Kommunikationsoplysninger, Kontaktdata, Andre oplysninger, du giver, Logdata, Brugsdata, Enhedsoplysninger, Lokationsoplysninger, Cookies og lignende teknologier.
Vi deler dine personoplysninger med leverandører og tjenesteudbydere, herunder udbydere af hostingtjenester, cloudtjenester, indholdsleveringstjenester, support- og sikkerhedstjenester, e-mail-kommunikationssoftware, webanalysetjenester, betalings- og transaktionsbehandlere.
OpenAI behandler dine personoplysninger på servere, der er placeret uden for EØS, Schweiz og Storbritannien... Når vi overfører personoplysninger uden for EØS, Schweiz eller Storbritannien, benytter vi gyldige overførselsmekanismer for at overholde gældende databeskyttelseslovgivning, såsom Europa-Kommissionens afgørelser om tilstrækkeligheden af databeskyttelse og standardkontraktbestemmelserne (SCC).
Vi bruger det indhold, du giver os, til at forbedre vores tjenester, for eksempel til at træne de modeller, der driver ChatGPT. Læs vores vejledning om, hvordan du kan fravælge, at vi bruger dit indhold til at træne vores modeller.
Vi opbevarer kun dine personoplysninger, så længe vi har brug for dem til at levere vores tjenester til dig, eller til andre legitime forretningsformål såsom løsning af tvister, af sikkerheds- og sikkerhedsmæssige årsager eller for at overholde vores retlige forpligtelser.
Missing or Unclear
- No named list of subprocessors (vendors/service providers) with roles and locations
- No detail on supplementary technical measures for international transfers (e.g., encryption, pseudonymization)
- No explicit restriction on data retention periods for non-critical data (e.g., logs, analytics)
- No clarity on whether ‘legitimate interests’ for model training meets GDPR’s high bar for consent in AI contexts
- No disclosure of data subject to US government requests (e.g., FISA 702) or how such requests are handled
- No information on Binding Corporate Rules (BCRs) or other transfer mechanisms beyond SCCs
Questions to Ask
- Can OpenAI provide a full, up-to-date list of all subprocessors (including sub-subprocessors) involved in processing EU user data, with their locations and roles?
- What supplementary technical measures (e.g., end-to-end encryption, zero-knowledge proofs) does OpenAI implement to protect EU data transferred to the US from government surveillance?
- Does OpenAI obtain explicit consent (not just an opt-out) for using user content to train models, given the high-risk nature of generative AI under GDPR?
- How long does OpenAI retain log data, usage analytics, and device identifiers for users who are not under active investigation or legal hold?
- What specific legal bases apply to the collection of location data, contact data, and device identifiers, and how are these justified as necessary?
- How does OpenAI ensure that vendors/service providers (e.g., cloud providers) do not access or use EU user data for their own purposes (e.g., model training)?
- Has OpenAI conducted a Data Protection Impact Assessment (DPIA) for its model training practices, and can it be shared with regulators or users upon request?
Share this analysis
Anyone with this link can view the result above.
Built by DentroChat
100% European AI chat for everyone
Chat with AI, work with files, generate images, and search the web. Data stays in Europe.
Other analysed privacy policies
View allrobotstxt.es
68/100
Medium privacy riskRobotstxt generally respects EU privacy but lacks clear details on legal bases, third‑party processors and AI use.
View report →fenritec.eu
68/100
Medium privacy riskFenritec generally respects EU privacy rules but lacks clear limits on data use for AI, proactive sub‑processor disclosure and detailed data‑minimisation statements.
View report →bestalternatives.eu
65/100
Medium privacy riskBest European Alternatives genuinely minimizes data and avoids cookies, but uses US-based processors without addressing international transfer safeguards and omits key GDPR rights and legal details.
View report →calendly.com
65/100
Medium privacy riskCalendly collects a wide range of personal and usage data, shares it with advertising partners, and while it offers standard EU rights and transfer mechanisms, its vague retention periods and silence on AI training for its Notetaker feature are concerning.
View report →