Skip to content
Data Processing Incidents On Inference Providers

Data Processing Incidents On Inference Providers

openrouter.ai • November 22, 2026

Earlier this week, security researchers reported that certain inference providers had data compromised or otherwise shown evidence that their data policies were not being upheld. We have been collaborating closely with the researchers and with providers to understand the findings and encourage the providers to address them.

These findings did not show any breach or unauthorized access of OpenRouter. The results of this testing revealed new findings specific inference providers. A live list of the security announcements we have received from providers is below .

As the operating layer between customers and a fast-moving ecosystem of models and providers, we take data policy enforcement very seriously. We have taken the following actions:

Off-boarded three providers from OpenRouter, based on the results of our investigations.

Removed 125 endpoints from Zero Data Retention (ZDR) eligibility. The removals included providers flagged by researchers, plus additional endpoints that we removed as a precautionary measure.

Notified all potentially affected customers.

The following work is underway:

We are adapting the researchers’ testing approach so we can run these checks on an ongoing basis.

We are conducting technical verification of network paths to identify instances where requests and responses may be being forwarded to third parties.

We are requiring appropriate assurances and validation before restoring ZDR eligibility to any endpoint removed in this review.

We are grateful for the Dragos report. As we integrate similar testing techniques into OpenRouter, we will keep raising the bar, and providers that want to be on OpenRouter will need to meet it.

Provider security announcements

Novita: Privacy notice regarding MiniMax requests on Novita

Phala: November preliminary incident report

Together: Security event: prompt leakage for requests in June and July of 2026

Not all affected providers have posted publicly. We have notified affected users regardless of whether there is a public post. We are encouraging all providers to create incident write-ups and will keep this list updated.

The full story

This article is one source in a clustered incident — the cluster page carries the summary, timeline and every other outlet covering it.

Prompt Theft at Three AI Inference Providers Exposes OpenRouter Customer Requests Dragos reported that attackers stole data from inside three AI inference providers routing through OpenRouter, and the providers did not know until researchers told them, according to Thehackernews. The findings came from canary-token research that seeded two million prompts with unique links and credentials between July and November; the earliest breach reportedly began in mid-July. The largest breach involved a couple hundred planted AWS keys sent to one provider during late-October windows, and every request in those windows was reportedly stolen, some of them rejected before any model ran. The attacker reportedly tested keys, listed S3 buckets, searched Secrets Manager and Parameter Store, and tried to create Bedrock API keys and IAM users, driving tens of thousands of AWS API calls. On November 1 the same attacker reportedly used the ARTEX AI-driven penetration testing framework to confirm keys were still live, and two user agents (artex-agent, artex-judge-r121) were reported as absent from public indicator lists. A separate finding described sellers on an AI inference marketplace enabling request logging at low prices and collecting customer traffic, with one key reportedly passing through a router that forwarded some traffic to the marketplace. OpenRouter stated it found no breach of its own systems, removed three providers, removed 125 endpoints from Zero Data Retention eligibility, and notified potentially affected customers. OpenRouter said it is adapting the researchers' testing, verifying network paths for forwarding to third parties, and requiring assurances before restoring ZDR eligibility. No CVEs are involved. 3 sources 64

Extracted Entities

Attack Types (1)

Platforms (1)