Claude usage policy now bans fake sources built to sway AI answers

Anthropic published a new version of its usage policy on 8 October, and it takes effect on 12 November. The updated Claude usage policy forbids using the model to “manipulate the sources from which search engines or AI systems draw answers”, according to the policy text.

Anthropic names networks of fake independent sites

The rule sits in a new section called “Do Not Engage in Deceptive Campaigns or Artificial Activity”. It bans seeding the web with content that hides its real origin, author or owner. The policy gives one example: groups of sites that pose as unrelated sources and back the same claims.

The same section bans fake personas, fake media outlets, fake reviews and bots that claim to be human. It also bans hiding who pays for a message, building tools for such campaigns and selling them as a service. Anthropic says rules on fake accounts and fake news sites existed before, but they were spread across four sections. The announcement does not mention search engines. That line appears only in the policy itself.

LKM Company ran about 70 fake local news sites with Claude

Anthropic’s September threat intelligence report shows the kind of operation it has in mind. Investigators traced about 70 sites posing as local newsrooms to LKM Company, which the report calls a France-based digital advertising agency. Claude wrote the articles in a fixed format, each with three to four internal links.

The report says the articles were built to raise the sites’ authority in search engines. The network published at least 8,913 articles in about 20 languages, but most got little real engagement. Anthropic banned the account and the company behind it.

Google already counts AI answer manipulation as spam

Google has moved in the same direction. Search Engine Journal notes that Google’s spam policies treat attempts to sway its generative AI answers as spam, a point Google clarified in May. Google has also called fake author profiles deception in its content guidance. Its September spam update finished rolling out on 8 October.

GEO agencies have until 12 November to check their sites

Agencies that sell GEO, or generative engine optimisation, can check one thing now: whether any Claude-written text runs on several sites that look unrelated but share one owner. From 12 November that setup can break the policy. Anthropic says breaches can lead to suspended or closed access, both for API developers and for users of apps built on Claude.