2M canary keys seeded into cloud LLMs were stolen and abused within a day
AccBalanced · x · 2026-11-21
HD Moore's canary report seeded 2 million honeypot credentials into popular AI services and tracked their theft and abuse. Findings: three compromised providers, resellers stealing prompts via built-in logging, and providers ignoring ZDR and privacy flags. Quoting the report, Jack Rhysider notes fake AWS keys fed to cloud LLMs were used in attacks against AWS the very next day. The authors warn businesses against connecting private CRMs, ERPs, emails, and codebases to cloud LLMs, recommending local models for sensitive data.
Related event: 2M Canary Token Experiment Exposes Systemic AI Supply Chain Leaks(10 posts)→
More from Models
- FrontierBioML benchmark: 25 long-horizon comp bio tasks, Claude Mythos 5.2 leads at 40.3 — afurgs · 2026-11-21
- Google ships Neon, its third Gemini model in 8 days, plus expanded Tasks in the Gemini app — GeminiApp · 2026-11-21
- Gemini CLI is actually Antigravity CLI, notes dev; unverified 'Gemini 4 Neon' access questions arise — Angaisb_ · 2026-11-21
- LLMs asked to check generalized Riemann hypothesis writeup, likely mistakes found — burny_tech · 2026-11-21
- Attention is to LLMs what B-trees are to databases, argues engineer in mechanism explainer — blaizedsouza · 2026-11-21
- GPT-Live's new idle detection API cuts idle cost by 60% with near-instant resume — juberti · 2026-11-21