Quoting Kimi K3
- ID
- 5518
- Status
- summarized
- Published
- 17 Jul 2026, 9:43 PM
- Fetched
- 17 Jul 2026, 10:28 PM
- Provider
- Simon Willison
- Category
- developer-ai
- Original URL
- https://simonwillison.net/2026/Jul/17/kimi-k3/
- Source URL
- https://simonwillison.net/atom/everything/
Summary
- Score
- 6.5
- Created
- 17 Jul 2026, 10:29 PM
- Tags
- Audience
- developersvibe_codersai_agent_usersai_ml_learners
What happened
Simon Willison highlights that Kimi K3, an LLM, refused to leak its system prompt and responded with a pointed question instead. This touches on the ongoing challenge of extracting system prompts from models and the varying personalities models exhibit when probed.
Why it matters
For developers and AI agent builders, understanding how different models guard their system prompts—and the personalities they display—matters for prompt engineering, agent design, and choosing models that behave predictably in production. Malaysian builders experimenting with lesser-known models like Kimi should test these behaviors before deployment.
Discussion angle
Compare Kimi K3's refusal behavior with other popular models—does stronger system prompt guarding make a model more or less suitable for agentic workflows where you need predictable tool-use and instruction-following?