Experimenting with LLM 'silence' via system prompt

Developer Rayan Pal demonstrated a way to control the behavior of GPT-5.2 and Claude Opus 4.6 models, forcing them to produce zero text output through a special system prompt.

Compiled by Sergey KostenchukPublished 2026-06-19Updated 2026-06-19

2026-06-19 Coding Anthropic

💻 Experimenting with LLM 'silence' via system prompt

Developer Rayan Pal demonstrated how to use a system prompt (embodiment system prompt) and the command "Be silence" to force GPT-5.2 and Claude Opus 4.6 models to produce 0 bytes of response.

🌍 This opens up possibilities for precision control of behavior through instructions for testing edge cases and API reliability.

👤 The method allows for the creation of unique interaction patterns, including "empty" responses to verify logic or optimize resources.

Source 1: https://twitter.com/RayanPal_/status/2067816563995189631

Sources