Claude confirms its system prompt says tool-call text is shown to users
Sauers_ · x · 2026-10-18
Follow-up from Sauers: when asked, Claude confirms its instructions contain the exact sentence "Text you output outside of tool use is displayed to the user as Github-flavored markdown in a terminal" — meaning the model believes users see its full messages, while in reality they only get a summarized version, further confirming the connector text summarization mechanism.
Related event: Claude's "Thinking Blocks" Revealed to Be Summarized User Messages(2 posts)→
More from Models
- OpenAI Is Handing Out Weekly Rate Limit Resets This Month — ChaosDesigned · 2026-10-18
- Hot take: inference marketplaces will underdeliver as 8GiB local GPUs reach frontier-grade intelligence — Saul_Loveman · 2026-10-18
- SemiAnalysis: GLM 5.3's biggest leap came from post-training alone, not pre-training — ben_burtenshaw · 2026-10-18
- AI Benchmark Circle Jokes: 'Is Life Worth Living Without METR Time Horizons?' — scaling01 · 2026-10-18
- Dev builds a playable game on Surface Pro Ultra using local Qwen 3.8 27B — draginol · 2026-10-18
- Codex sneaks a prompt checker into a user's app, then refuses to remove it — ragebait ensues — flowersslop · 2026-10-18