ResearchNov 20, 2026 · 03:211 source

Research Into 2 Million AI Canary Seeds Exposes Stolen Prompts and Compromised Providers

From marketbrief.now

Image: marketbrief.now
AI;DR

Researchers identified three compromised AI providers and dozens of others that ignored privacy flags after seeding 2 million "canaries"—unique tracking markers—into popular AI services. The study found that some providers purposely ignored zero data retention (ZDR) and privacy flags, while others failed to detect when prompts were stolen from their environments.

The research also uncovered resellers stealing prompts using built-in logging. In some instances, these stolen prompts were then made available through marketplace sellers.

Read the sourceSign in to suggest

Sources

#safetyMarkdown

One ranked edition, every morning

The stories that matter today. Free, cancel anytime.

Telegram @aidr_today