Back to Moonshot AI (Kimi)All frontier AI labs
Ethics and safety policies
Moonshot AI (Kimi): what they say, and what they leave out
Every claim on this page comes from documents Moonshot AI (Kimi) published itself, linked below so you can check me. Checked September 19, 2026.
A policy is not an audit
Almost every document linked on this page was written by the company it governs, and graded by that same company. A long policy list does not make a lab more ethical than a short one. This page keeps three things apart: what the lab has committed to in writing, what its documents do not cover or where they have been credibly criticized, and — clearly labeled at the end — my own read.
Their stated position
Moonshot publishes strong open-weight models with technical reports, under its own license rather than a standard open one, and says relatively little publicly about safety policy.
Its hosted app operates under Chinese generative-AI regulation.
The documents themselves
Read these rather than this summary if a decision depends on it.
- Kimi model repositories and license
Weights, the Modified MIT license terms and published evaluations.
- Platform terms
Permitted use of the API and hosted app.
What they have committed to
Stated in their own published policies, in writing.
- — Publishes weights and technical reports for its flagship models.
- — Documents license conditions explicitly rather than leaving terms implicit.
What those documents don't cover
Plain absences, and criticism that has been documented publicly — not speculation.
- — No frontier-risk framework, no system card practice, little published red-teaming.
- — Its license adds attribution conditions above a standard MIT license, so 'open' needs reading carefully.
- — The free app tier runs an older model than the open weights, which complicates any claim about what 'Kimi' does.
My read
This section is my opinion, clearly separated from everything above. Disagree with it freely.
Worth using and worth reading the license for. Safety governance is essentially undocumented, so treat outputs as unvetted.