Anthropic's new usage policy bans using Claude to seed AI answers with hidden-origin content
An update announced on 8 October and effective 12 November adds search engines and AI tools to Claude's prohibited uses.
Lily Ray flagged a catch by Hanns Kronenberg and Matt G. Southern on LinkedIn on 11 October: Anthropic updated its Usage Policy on 8 October, with the new terms taking effect on 12 November.
A new section bans using Claude to seed the sources that search engines and AI tools draw answers from with content that hides who wrote it or who stands behind it. Anthropic's example is networks of sites posing as independent sources that back the same claim.
The policy also bans pushing content through networks of fake or supposedly independent sites to hide that it comes from one source. Building or selling tools that do any of this is prohibited as well. The penalty is losing access to Claude.
Ray notes the previous version, from September 2025, did not mention search engines or AI answers at all. Fake reviews and fake personas were already banned under the earlier terms.
What the rule actually covers
The scope here is use of Claude itself, enforced through account access. Anthropic cannot remove pages from the open web or from any retrieval index. What it can do is cut off the tool used to generate coordinated material at scale, and it has now written that intent into policy.
For practitioners, we think the practical read is narrower than a general crackdown on AI-written content. The named harms are concealed authorship and coordinated inauthentic networks, which were already risky ground under platform rules and under Google's spam policies.
How to check your own exposure
Before changing anything, audit what you are actually running. List every property, persona and syndication partner that carries content generated with Claude, and check whether a reader could identify the party behind each one. Where authorship is obscured or where several outlets you control echo the same claim, that is the pattern the policy names.
Fix the disclosure and the coordination first. The underlying question of whether a page deserves to be cited by an AI answer sits upstream of any single model's terms, and that is where sustained work pays off regardless of what Anthropic enforces in November.