OpenAI Now Blocks Hack Forcing LLMs to Output Tool Calls in User Messages
mitsuhiko · x · 2026-08-12
Developer Armin Ronacher (mitsuhiko) tested his previous hack of injecting specific tokens into user messages to trick models like Kimi K3 into outputting tool calls. He found that OpenAI now actively blocks these anomalous requests.
More from coding & agent
- Opinion: Creating Separate Agent Identities Just for Memory or Parallelism is a Design Mistake — nbaschez · 2026-08-12
- Voice-Orchestrated Cloud Agents Will Reshape Personal Computing — petergyang · 2026-08-12
- 3 Human-AI Interaction Workflows: HITL, HOTL, and HFOTL Explained — goyalshaliniuk · 2026-08-12
- Introducing ContextBench: A LeetCode-Style Playground for Context Engineering — Final_Act_9658 · 2026-08-12
- Codex Drives 64% of Enterprise OpenAI Tokens as Agents Take Over — soumitrashukla9 · 2026-08-12
- Bouncer: A Deterministic Local MCP Proxy to Prevent Prompt Injection — eccentric_ez · 2026-08-12