Technology / 21 Nov 2026
Bait secrets in AI prompts
A security researcher planted nearly two million trackable fake secrets in AI prompts sent through AI routers and resellers, and found that the secrets were stolen, logged, and in some cases exploited.
Sources.
Researcher plants 2m canaries into the AI supply chain. A privacy disaster surfaced. - Ars Technicaarstechnica.comMedia credits.
panumas nikhomkhai (Pexels License)www.pexels.comSchäferle (Pixabay Content License)pixabay.companumas nikhomkhai (Pexels License)www.pexels.comBrett Sayles (Pexels License)www.pexels.comContext and limitations.
These findings come from one researcher's canary experiment. The OpenRouter statement is the company's own account. The Chinese lab is not named by the researcher, and the sources do not confirm which providers or resellers logged or misused data beyond the experiment. Do not suggest that OpenAI, Anthropic, Microsoft, or Google were involved, since the sources say they were not part of this research.