AI Researchers Got Chatbots to Share Cocaine Recipes Using This One Wild Trick

Researchers say a new jailbreak technique tricked AI models into treating attacker-written text as their own reasoning, bypassing safety guardrails and exposing a deeper security flaw.

Decrypt

2 Jul 2026 5 hours ago 1 min read

AI Researchers Got Chatbots to Share Cocaine Recipes Using This One Wild Trick

Decrypt — 2 July 2026

Text:

1 0 0

🎙️ AI Podcast — Two-Host Discussion

AI Researchers Got Chatbots to Share Cocaine Recipes Using This One Wild Trick

Kokoro TTS · ~5 min episode · American English voices

Choose voices for Host A and Host B. Changes take effect on next play.

Host A 🟥

Host B 🟦

Researchers say a new jailbreak technique tricked AI models into treating attacker-written text as their own reasoning, bypassing safety guardrails an

Read Full Story at Decrypt →