2 Million Bait Requests Reveal Data Leak Risks in AI Service Chains
Security firm Dragos's chief research officer HD Moore sent nearly 2 million requests with trackable "canaries" to multiple AI providers to trace where data actually goes.
The team gave AI trackable test secrets and watched whether other programs or people used them to access systems. At least three providers showed signs of a breach, with attackers using stolen bait keys for tens of thousands of cloud API calls. Some keys were used within hours of being submitted.
Risk also comes from layered supply chains. Some resellers recorded and shared requests, and private URLs marked "zero data retention" were still scraped days later.
Moore suggests local deployment for sensitive work, or choosing trusted providers directly and naming the upstream provider.
Comments
Loading comments…