UK AISI and CAISI publish first joint cyber assessment of Kimi K3

The NIST website hosts a preliminary joint assessment by UK AI Safety Institute and CAISI (US) of Kimi K3's cyber capabilities - the first formal cross-Atlantic evaluation of a Chinese open-weight frontier model.

The fact

On July 25, NIST published a joint preliminary evaluation of Kimi K3's cyber capabilities, signed by UK AISI (AI Safety Institute) and CAISI (Center for AI Standards and Innovation, US side). The announcement is short and points to a report hosted on nist.gov. This is the first time these two institutions have publicly co-signed on a Chinese open-weight model.

Our analysis

The political signal is stronger than the technical content available at this stage. Kimi K3, released in preview and then live in July (thread kimi-k3-launch), was immediately analyzed in the US as a cyber risk - publication #1591 reported the "AI communism" reading from Wall Street. The fact that AISI and CAISI publish a formal evaluation - even preliminary - instead of a classified briefing sets a standard: foreign frontier models are now evaluated publicly, similar to what AISI did for Anthropic and OpenAI models over the past two years.

To watch

• The Chinese reading: official responses from Moonshot or MIIT.
• The shift: will the "AISI + CAISI joint report" format become standard for each new Chinese frontier model (Qwen 3.8 Max, next DeepSeek)?
• The technical content of the preliminary evaluation - exploit generation capability, network recognition - as soon as it is indexable.