I had some Custom Assistants set to GPT 5.1 and to Mistral Medium.
These models were replaced with GPT 5.2 and Mistral Large 3.
However, the Custom Assistants became set to Ki Quick instead (which is Claude under the hood?).
This might not be so great if one doesn't notice first, especially if concerned about privacy with Claude - https://help.kagi.com/kagi/ai/llms-privacy.html indicates they might retain prompts for 2 years if they're deemed naughty?
Either:
- Automatically switch to the closest fit from the same model provider
- Throw an error until the Custom Assistant is manually updated
- Switch to whatever is the most "private" LLM