Skip to main content
Question

AI Watermarking with Anthropic LLMs

  • September 4, 2026
  • 2 replies
  • 20 views

Matt Hoey
Forum|alt.badge.img

With Anthropic having deployed watermark technology applied to responses from their LLMs, does this invisible green/red watermarking also apply to custom agent outputs in Box when we assign an Anthropic LLM to the agent?

  • Example 1 -  we assign Opus 5 to a specialist workflow agent in Box AI Studio and this agent is selected by a user to query against a file. Will that response output include watermarking?
  • Example 2 - we use the same agent as above but this time we trigger it to perform an action as part of a Box Automate flow. Will that response output include watermarking?

This video (no affiliation) was useful to me to learn about AI watermarking.

2 replies

Matt Hoey
Forum|alt.badge.img
  • Author
  • New Participant
  • September 4, 2026

Correction to my examples above, I believe Anthropic models are watermarking beginning with Fable 5.1 and going forward so same questions but assume Fable 5.1 or newer is assigned to the agents.


Jey Bueno Box
Forum|alt.badge.img
  • Community Manager
  • September 5, 2026

Hi ​@Matt Hoey,


Thank you for reaching out with this question regarding AI watermarking and custom agent outputs.


We are actively collaborating with our LLM partners, including Anthropic, on how inline watermarking of content will be implemented across our AI integrations (such as Box AI Studio agent queries and Box Automate flows). Box is committed to ensuring full compliance with EMEA regulations and relevant regional standards, and we are currently working through the final implementation details with our partners. We anticipate finalizing these details as part of our Q4 activities.


We will share further updates as these capabilities are rolled out. Thank you!