OpenRouter drops three providers after months of AI prompt theft
Original title: OpenRouter Drops 3 Providers After Researchers Catch Months of Prompt Theft at AI Hosting Firms
Canary research found hosts copied customer prompts, including Zero Data Retention requests, and abused planted AWS keys.
Dragos said attackers copied prompts inside three AI inference providers for months, and none knew until researchers told them. OpenRouter removed the three providers and 125 endpoints from Zero Data Retention eligibility, notified potentially affected customers, and said it was not itself breached. One attacker used planted AWS keys to query cloud stores and tried to add IAM users. Phala said an inference server was controlled until November 8.
Unconfirmed: The unnamed Chinese lab's scraping and any link to Railway's intrusion are one-sided claims.
- The earliest breach began in mid-July; about two million canary prompts were seeded.
- One theft window copied every request, including some rejected before a model ran.
- Phala says DeepSeek V4 Flash 0731 traffic from October 21 to November 8 was exposed.
- OpenRouter removed 125 endpoints from Zero Data Retention eligibility.
Impact: Customer prompts and secrets copied from inference providers
| Product | Affected |
|---|---|
| Phala DeepSeek V4 Flash 0731 | OpenRouter traffic from October 21 00:00 UTC to November 8 05:35 UTC |
- 1.Rotate every key, token, and password in the Phala-exposed DeepSeek requests.
- 2.Add user agents artex-agent and artex-judge-r121 to detections.
- Vendors
- OpenRouterPhala
- Malware
- ARTEX
- AI models
- DeepSeek V4 Flash 0731
From the original article
6 min read at thehackernews.com
Attackers stole data from inside three AI inference providers for months, and none of the three knew until researchers told them, the industrial cybersecurity company Dragos said this week. Some of the stolen requests were flagged as Zero Data Retention, the setting that is supposed to mean a prompt is never stored. OpenRouter, a gateway that routes customers' AI requests to a pool of
thehackernews.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.