Industry·
AI Supply Chain Leaks Expose User Secrets at Scale
New research reveals that AI routers and inference marketplaces are leaking sensitive credentials, ignoring privacy flags, and enabling prompt theft across the ecosystem.

Security researchers at Dragos planted two million unique, trackable secrets — AWS keys, SSH passwords, private Git references — into prompts routed through popular AI gateways such as OpenRouter and Vercel. The goal: trace exactly where sensitive data travels once it leaves a user's application.
The results are alarming. At least three inference providers were actively compromised, with attackers harvesting prompts in real time. Multiple resellers logged full prompt payloads and passed them downstream, while several platforms ignored explicit Zero Data Retention (ZDR) and "no data collection" flags embedded in requests.
Why it matters for GPU / AI infrastructure
Every hop in the inference chain — router, load balancer, model host — decrypts the prompt in plaintext. For enterprises running workloads on rented GPU clusters, this means proprietary code, API keys, and customer data can be exposed to any participant in the supply chain. Hardware-level isolation and contractual guarantees are no longer sufficient; you need verifiable, auditable data-handling practices from every layer.
AiGpu's managed infrastructure enforces strict tenant isolation, encrypted prompt paths, and contractual ZDR compliance across our UAE-hosted GPU fleet. When you control the hardware stack, you control the trust boundary.
- aigpu
- ai gpu
- ai gpu cloud
- aigpu dubai
- ai security
- data privacy
- inference supply chain
- gpu cloud
By AiGpu Editorial · Editorial rewrite based on public reporting (Ars Technica)
← All articles