Dragos planted fake AWS keys in 2M AI prompts — the keys got used
serendip-ml · reddit · 2026-11-22
Security firm Dragos ran a canary-token test of AI supply-chain privacy: from July to November they sent 2 million prompts through AI providers, routers, and reseller marketplaces, each carrying a unique fake AWS key or secret link.
Findings
- Multiple providers had been breached and didn't know until Dragos told them
- Fake keys were used for tens of thousands of AWS calls, some from prompts rejected before a model even ran
- Resellers could read and retain customer prompts via ordinary logging
- The "don't retain my data" flag didn't help
Takeaway: prompts pass through more hands than the paid provider; the only prompt that provably stays home is one answered by your own GPU.
More from Infra
- Nemotron 3.6 Ac Academia Support Lands in llama.cpp: Looped Hybrid with Gated Delta Product — maternion · 2026-11-22
- Sparse MoEs make SSD the last bottleneck: expert streaming breaks the VRAM wall — rickasaurus · 2026-11-22
- Inkling 2 Mini on 2× RTX 3090: 557K Context + Vision at ~100 tok/s via NInfer-all — Sisuuu · 2026-11-22
- OpenAI engineer confirms GPT-Live cache affinity bug, fix rolling out Monday — pbbakkum · 2026-11-22
- Lambda runs Nvidia CPO switches on 10,000+ GPU GB300 NVL72 cluster, cutting latency up to ~10% — Beth_Kindig · 2026-11-22
- Running Inkling 2 Mini locally on Mac: 20GB model, ~16 tok/s on M4 Pro — jacob_posel · 2026-11-22