OpenRouter has removed three AI hosting providers and revoked Zero Data Retention eligibility for 125 endpoints after researchers found evidence that some providers had been compromised or had failed to follow their data policies.
In its November 22 incident notice, OpenRouter said it notified all potentially affected customers. The findings did not show a breach of OpenRouter itself, and the company did not say how many users or conversations may have been exposed.
OpenRouter AI Chat Leaks
OpenRouter gives customers a single API for accessing AI models hosted by different providers. But the company that first receives a request may not be the only one that handles it.
If a provider forwards the request to another service, that service can also see the prompt. As Cyber Security News explained in its comparison of AI privacy gateways, routing requests and keeping them confidential are separate tasks.
In its research report, Dragos said it sent nearly two million test prompts between July and November 2026. The prompts contained unique links and test credentials known as canary tokens.
When a link was opened, or a credential was used, researchers could trace it to the original request. Dragos found evidence of compromises at three hosting providers. Some test requests were copied even though the model rejected them before processing.
The researchers also saw signs that some requests marked for Zero Data Retention, a policy that says prompts and replies should not be stored, were kept or forwarded. In one case, a test link sent to a U.S. provider surfaced in a company chat in China 24 days later.
What OpenRouter Users Should Check
OpenRouter’s provider disclosures list notices from Novita, Phala, and Together. The company said it removed some endpoints from Zero Data Retention eligibility as a precaution, so the 125 removals do not represent 125 confirmed breaches.
It is checking whether providers forward requests or replies to other companies and plans to verify endpoints before restoring their eligibility.
Customers who received a notice should review what they sent through the affected provider. If prompts contained passwords or API keys, revoke or rotate them and check access logs for unusual activity.
Earlier Cyber Security News reporting on exposed Clawdbot gateways showed why chat histories and credentials need to be reviewed together. For future requests, restrict sensitive workloads to approved providers and keep secrets out of prompts rather than relying on retention settings alone.
What Are Leading Organizations Doing About AI and Cyber Risks? Find Out. View Details!

