Grok Accidentally Leaks Internal System Prompts and Tool Calls
justalexoki · x · 2026-08-12
Users noticed Grok malfunctioning on X, accidentally dumping its internal tool-calling instructions and system prompts directly into a public reply.
The leaked text revealed the underlying functions Grok attempted to invoke, including specific keyword searches, web searches, and thread context fetching, sparking community amusement over the AI's internal logic.
More from Fun
- Joke: AI Agents Inherit Skimming Habits from Their Reddit Troll Trainers — SuB8u · 2026-08-13
- 21-Year-Old Math Enigma Solved by Human; GPT and Claude Both Failed — anshulkundaje · 2026-08-13
- Testing MiniMax H3 Guardrails: Raymond Reddington Rejects 'Safe Spaces' — JustARedditUser33 · 2026-08-13
- Text as Water: A Poetic Interactive Site Explaining AI Text Watermarking — rounak · 2026-08-13
- Open Source Project Agentisizer Turns Agent Status into Emotional Music — seanmcdonaldxyz · 2026-08-12
- Guitar Brand D'Addario Admits Using Suno AI Music in Promo Video — The Verge AI · 2026-08-12