Skip to content
NewA calmer, faster ZeroHour

Dragos finds compromised AI providers harvesting AWS keys from prompts

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

The Decoderby Matthias Bastian ()3 min read1
Part of a story covered by 4 sourcesOpenRouter drops three hosts after canary tests find copied promptsMerged summary & timeline
AI summary · x-ai/grok-4.7Summaries by Vani AI

OpenRouter removed three providers and revoked zero-data-retention approval for 125 endpoints after the July–November tests.

Dragos sent about 2 million requests containing fake credentials between July and November 2026 and found at least three providers already compromised. Attackers harvested hundreds of AWS keys, including from rejected requests, then tested keys, searched S3, and sought persistent access. Resellers could read prompts without a vulnerability by offering the lowest price and logging traffic, and zero-data-retention settings were often ignored. Novita, Phala, and Together published preliminary reports.

Unconfirmed: Whether the examined providers share personnel with Chinese proxy networks is still unclear.

Key facts
  • Tests ran from July through November 2026 and sent about 2 million requests with fake credentials.
  • The largest case harvested hundreds of AWS keys, including keys from rejected requests.
  • A French residential connection tried SSH logins with passwords taken from intercepted requests.
  • A secret test link appeared in a Feishu chat 24 days later and was opened in China.
  • A Chinese AI lab reached secret test URLs in more than half of requests, Dragos says.
Security assessment · AI, from the article

Impact: AWS keys and prompts exposed; stolen keys used on S3

ProductAffected
Novita—
Phala—
Together—
What to do
  1. 1.Do not send requests containing credentials to AI intermediaries.
  2. 2.Control how AI requests are routed and review privacy settings.
  3. 3.Plant canary tokens that alert when fake credentials are used.
Explain this AI news with Vani AIWhat was announced, what is new and why it matters, in a new chat

From the original article

3 min read at the-decoder.com

Security researchers at Dragos used two million fake AI requests to uncover serious vulnerabilities in the AI supply chain, finding at least three compromised providers, hundreds of stolen AWS keys, and marketplace resellers who were simply reading customer requests. OpenRouter responded by removing three providers and 125 endpoints.

The Decoder

By Matthias Bastian · the-decoder.com

Read the full article at The Decoder

A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.