“When a chat turns into a statement”: OpenAI scans ChatGPT conversations and may contact the police — what this means for the privacy of Ukrainians
- Максим Бадзюнь
- Aug 29
- 3 min read
OpenAI has officially explained that detected chats with signs of planning to harm others go through a separate moderation route: they are reviewed by a small team that can ban the account and refer the case to law enforcement if there is an "imminent threat of serious physical harm." This is stated in a recent clarification from the company. OpenAI
The news was picked up by the media: Futurism describes the solution as a more explicit statement that OpenAI scans conversations and is able to notify the police — and this has already caused a wave of reactions from users.
What is the legal basis for this?
In its global privacy policy, OpenAI explicitly states that it may share personal data with government agencies and other third parties: to enforce the law, prevent fraud, protect the safety of users and the public , and in case of violation of the terms of use.
Separately, OpenAI emphasizes usage policies : prohibiting illegal activities, compromising the privacy of others, etc. Such policies provide grounds for content moderation and data transfer in the event of threats.
What exactly is being “scanned” and when do people intervene?
According to the company, automatic risk signals (topics of violence, planning to harm others, etc.) direct conversations into a separate pipeline; then human reviewers assess the context and make decisions — from warning and blocking to notifying the police in the case of a real and imminent threat.
Why is this controversial?
Critics note that chatbots often collect very sensitive information (therapy, legal issues), while lacking the confidentiality of a doctor-patient/lawyer-client level . Such services can become an element of a “surveillance” infrastructure if there is no encryption and clear access limits.
How does this relate to user rights in Ukraine and the EU?
GDPR/Ukrainian law on personal data protection (very similar in spirit): allows processing and transfer of data on a legal basis (legal obligation, vital interests, legitimate interests). Transfer “for the protection of life/safety” is a typical legal basis, but requires proportionality and data minimization .
Users have rights of access/deletion/restriction of processing; however, in the event of investigations or legal obligations, these rights may be temporarily restricted . (This block is general information, not legal advice.)
What really changes for the average user
Open threats to other people in the chat are now more likely to be subject to human moderation and may be passed on to law enforcement . This is an officially declared procedure.
OpenAI's policy has long included the ability to share data with government agencies as required by law; now the company has more clearly described the operational mechanism for such a route.
AI Academy Privacy Checklist (what to do today)
Do not share real plans in the chat that could be interpreted as a threat or illegal action — even "as a joke."
Enable enhanced privacy modes : temporary chats / history off (and, if available in your plan, Zero Data Retention ). Check your corporate settings if you use it at work.
Minimize personal data : names, addresses, numbers, medical/legal details.
Local or on-prem solutions : For particularly sensitive cases, consider local models or contractual terms with the supplier at the DPA/SCC level.
Stay tuned for updates on OpenAI policies (retention, encryption, exceptions for law enforcement).
A brief “data map” of OpenAI (as of today)
Table - short paragraphs, no long sentences.
What | How/why | Transfer to the authorities |
Metadata and interaction | security, abuse prevention | possible by law/threat |
Chat content | moderation, service improvement (depends on settings) | in the event of an "imminent threat" |
Temporary chats | short storage, separate pipelines | by threat/law |
Enterprise | limited access, ZDR options | upon legal request |
Sources: privacy policy, usage-policies, OpenAI clarifications.OpenAI+3OpenAI+3OpenAI+3
Academy Conclusion
From now on, it is transparently recorded : if ChatGPT records a real threat to others , the conversation may end up with law enforcement. This increases public safety , but at the same time raises the bar for privacy : clear boundaries, encryption, understandable retention, and an independent moderation audit are needed. Users need digital hygiene and thoughtful settings; businesses need compliance and contractual guarantees ; the state needs proportionality rules and transparency of data access.
Comments