Breaking
OpenRouter Off-Boards Providers, Together AI Reports Prompt Leakage After Data Incidents
OpenRouter has removed three inference providers and 125 endpoints from zero data retention eligibility following security incidents, while Together AI confirmed prompt leakage for requests in June and July 2026.

By Elena Rostova, Cybersecurity
4 min read
OpenRouter, an AI inference router, has taken significant steps to enhance data security after reports from security researchers indicated data compromises or policy breaches at several of its inference providers. The company announced it has off-boarded three providers and removed 125 endpoints from its Zero Data Retention (ZDR) eligibility list.
One of the affected providers, Together AI, publicly disclosed a security event involving "prompt leakage for requests in June and July of 2026." This incident highlights the critical privacy concerns associated with third-party AI inference services, where user prompts and data could be inadvertently exposed.
What Happened?
Security researchers identified vulnerabilities or policy violations at various AI inference providers. These findings prompted OpenRouter to conduct its own investigations and collaborate with affected providers. The incidents did not involve any breach of OpenRouter's own infrastructure but rather issues at the provider level.
OpenRouter's Zero Data Retention (ZDR) programme requires participating providers to meet strict data handling standards. When researchers reported that certain providers had experienced data compromises or failed to uphold their ZDR commitments, OpenRouter acted swiftly to remove them from the programme.
OpenRouter's Response
In its announcement, OpenRouter stated it has:
Off-boarded three inference providers from its platform
Removed 125 endpoints from ZDR eligibility
Notified affected customers directly
Implemented continuous testing and technical verification of network paths
Encouraged providers to publish their own incident write-ups
The company emphasised that these actions are part of an ongoing effort to maintain trust and transparency in the AI inference ecosystem.
Together AI's Disclosure
Together AI, one of the affected providers, published its own security event disclosure confirming "prompt leakage for requests in June and July of 2026." The company stated that the issue has been resolved and affected customers have been notified. Together AI's transparency in publishing its own write-up aligns with OpenRouter's call for provider accountability.
Why This Matters
As AI inference becomes increasingly central to applications, the security of data flowing through third-party providers is paramount. Users and enterprises routing sensitive prompts through inference gateways need assurance that their data is handled according to stated policies. This incident demonstrates both the risks of the current model and the importance of independent verification.
What to Watch
Whether other affected providers publish their own incident reports
How OpenRouter's continuous testing framework evolves
Industry-wide adoption of technical verification for data retention claims
Potential regulatory attention to AI inference provider accountability
This story was first published by TechWhoKnows on 22 November 2026. Original: https://techwhoknows.com/security/openrouter-off-boards-providers-together-ai-reports-prompt-leakage-after-data-incidents
Sources
- 1OpenRouter announcement: Data Processing Incidents on Inference Providersopenrouter.ai
- 2Together AI blog: Security Event - Prompt Leakage for Requests in June and July of 2026together.ai
- 3Hacker News discussionnews.ycombinator.com