← Back to the wire

Security researchers show how cheap AI inference providers steal keys and snoop on users

AnnouncementProductNov 22, 2026

Dragos researchers sent roughly two million test requests with fake credentials to AI providers and found at least three compromised intermediaries stealing user data. In the largest incident, attackers harvested hundreds of AWS keys, one using ARTEX, an AI-powered attack tool. Dragos warns that marketplace intermediaries can read requests and recommends canary tokens, controlled routing, and checking privacy settings, noting that low-price routing decides where prompts go.

Receipt № 30241 source · awaiting confirmation ◐

Evidence

1source· awaiting independent confirmation

No score is assigned. Sources and their independence are shown in the citation chain below.

Citation chain · 1 source

01med
the-decoder.comBloghttps://the-decoder.com/security-researchers-show-how-cheap-ai-inference-providers-steal-keys-and-snoop-on-users/This source details the experiment's findings on compromised intermediaries, ignored retention settings, and OpenRouter's response, framing low-cost routing as the core security decision users fail to make.
PRIMARY
AWSCompanyARTEXModelDragosCompany
Canonical: https://the-decoder.com/security-researchers-show-how-cheap-ai-inference-providers-steal-keys-and-snoop-on-users/