AI Trends

Nearly 2 Million Decoy Requests Expose Privacy Risks in the AI Service Chain, With Leaks Possible Even Under Zero Data Retention Settings

IT Home · Artificial IntelligenceUpdated Nov 21 22:16

AI summary

Dragos Chief Research Officer HD Moore sent nearly 2 million requests containing traceable decoys to multiple AI service providers, uncovering issues including stolen prompts, data leaks by resellers, and failures to honor privacy settings. At least three providers showed signs of compromise, and some decoy keys were used to make tens of thousands of cloud service API calls within hours of submission. Private URLs in some requests marked for zero data retention or no data collection were still accessed days later.

Why read this

SelectedAI score 78

Nearly 2 million requests containing traceable decoys reveal how prompts, keys, and privacy settings can be exposed across multilayer AI service chains.

Topics

Tags

Body · AI translation

Nearly 2 Million Requests Expose AI Privacy Flaws: Even “Zero Data Retention” Can Leak Secrets

November 21, 2026, 10:03 — ITHome — Guyuan

ITHome reported on November 21 that technology outlet Ars Technica published a blog post today (November 21) stating that HD Moore, chief research officer at security company Dragos, sent nearly 2 million requests containing traceable “canaries” to multiple AI service providers and discovered problems across the AI service chain, including hackers stealing prompts, resellers leaking data, and privacy settings not being honored.

The researchers deliberately sent AI systems a batch of traceable “test secrets” (not collected through real-user channels), then watched to see whether other programs or people subsequently used them to access systems. This allowed them to determine whether the data had leaked and which part of the service chain it might have leaked from.

The investigation found signs that at least three service providers had been compromised. Attackers used stolen canary keys to make tens of thousands of cloud service API calls in attempts to find additional credentials, activate services, and maintain access. Some keys were used within hours of being submitted. In some cases, requests were rejected before the model even ran, yet the keys they contained were still stolen.

Risks also arise from service chains involving multiple intermediaries. Many AI applications route requests through brokers, and the content may pass through several platforms, where it is decrypted and processed at each stage. The research also found that some resellers logged and transmitted requests externally. Private URLs in some requests marked “zero data retention” or “data collection prohibited” were still crawled several days later.

Moore recommends considering local deployment for sensitive operations. Users unable to build their own systems should choose trusted service providers directly whenever possible and explicitly specify the upstream provider.

Reference link provided by ITHome

Three breaches, endless leaks: Your AI supply chain needs canaries

Advertising statement: External links contained in this article, including but not limited to hyperlinks, QR codes, and passcodes, are provided to convey additional information and reduce the time needed to evaluate options. The results are for reference only. All ITHome articles containing external links include this statement.

Source

IT Home · Artificial Intelligence