Hi Kagi team,
Following Anthropic's announcement of text watermarking and an RFC proposing native watermarking support in vLLM, could you clarify Kagi's position on watermarking AI outputs?
Text watermarking can allow generated text to be identified probabilistically after it leaves the service. My concern is that users could unknowingly receive text carrying that signal, even from a privacy-focused service, with wording potentially chosen not because it best serves the user, but because it makes the output easier to track.
Kagi's privacy policy addresses data handling, but doesn't appear to cover whether outputs from Assistant, Universal Summarizer, FastGPT, Translate, or other AI features are watermarked or otherwise marked for later detection.
Could you please:
An explicit policy would mean a lot to those of us who chose Kagi because we trust it not to introduce privacy-affecting changes quietly.
Thank you.
N/A