Third-Party AI Providers Exposed Private Chats by Ignoring Data Retention Policies
Written by AI from the linked story. Not reviewed by an editor. How this works
OpenRouter's reliance on third-party AI hosting providers created a supply chain trust failure: some providers either suffered their own compromise or deliberately disregarded Zero Data Retention (ZDR) commitments, logging and forwarding user prompts they were contractually obligated to discard. Critically, researchers found that prompts were copied even before the model processed them, meaning data left the user's control at the infrastructure layer regardless of model-level rejections. This incident illustrates that data protection guarantees are only as strong as the weakest third-party link in the service chain. Users who sent sensitive information — including credentials or API keys — through affected endpoints may now face credential theft or data exposure risks.
Tactical Insight
Immediate actions
- Revoke and rotate all API keys and passwords sent through affected OpenRouter endpoints as identified in vendor notifications.
- Audit logs of requests made through affected providers to determine the scope of potentially exposed data.
Long-term improvements
- Establish contractual audit rights and periodic technical verification (e.g., canary tokens, honeypot prompts) to confirm third-party data retention compliance.
- Require third-party AI providers to undergo independent security assessments before being granted Zero Data Retention eligibility.
- Implement a formal third-party risk management program with ongoing monitoring and clear offboarding procedures when providers are found non-compliant.
Detection measures
- Deploy automated testing that sends tagged synthetic prompts through providers and monitors for unexpected reappearance or exfiltration.
- Establish a continuous vendor risk scoring system that flags anomalies in provider behavior or security posture for rapid response.