Anthropic Just Made Fake-Sourcing Harder — And It Actually Matters

2026-10-11 — brand building small business India

Someone built 70 fake local news sites. The sites existed only to get quoted by AI systems.

That's the actual problem Anthropic is trying to solve.

Anthropic's September 2026 threat report described about 70 fake local news sites that published at least 8,913 articles in about 20 languages, built with Claude prompts that demanded three to four internal links per article. Not for humans. For AI crawlers. The whole infrastructure was upstream manufacturing.

Effective November 12, Anthropic prohibits using Claude to "manipulate the sources from which search engines or AI systems draw answers by seeding them with content that misrepresents its origin, authorship, or independence."

The rule lives in a new section called "Do Not Engage in Deceptive Campaigns or Artificial Activity." It's specific.

What does that mean in practice? You can't run a network of sites that pretend to be independent but all say the same thing. You can't use Claude to build the tools that run those networks. You can't sell that service to someone else. And if Anthropic finds out you're doing it, they can cut off your access — warning you first, maybe, or just terminate the account.

The policy already banned fake accounts and fake news sites. But before October, those rules were scattered across the elections section, the fraud section, the privacy section, and disinformation guidance. Most consultants didn't read the whole thing, including us sometimes. The guidelines were previously scattered across the parts of its rules on elections, fraud, privacy, and disinformation.

Now they're consolidated.

Actually, that's not quite right — it's more than consolidation. The new language is explicit about AI answer sources. The 2025 version didn't mention search engines or AI responses at all. You could probably argue your way out of it. The 2026 version names the thing directly.

Anthropic's September 2026 threat report describes about 70 fake local news sites that published at least 8,913 articles in about 20 languages, built with Claude prompts that demanded three to four internal links per article. Someone was doing this at scale. That's why Anthropic updated the policy.

The rule also covers something slightly different: fake personas, fake outlets used to hide who's actually behind the content, and building tools designed for deceptive campaigns. The common thread is authorship. If you're hiding who wrote something or who told you to write it, you're out.

One thing changed for the better, or at least softened. Automated publishing used to be on Anthropic's high-risk activities list. Now it's not. That matters if you're running a service that automatically generates content and publishes it somewhere. Automated publishing is no longer on the policy's list of high-risk uses. It's still covered by the broader policy if you're being deceptive, but it's not presumed dangerous anymore.

The policy binds Claude users, not search engines. Anthropic can't force Google or Perplexity to stop indexing fake sites. But it can tell you not to build them using its product.

That's the leverage.

Source: "From 12 November 2026, Anthropic's Usage Policy bars using Claude to seed search and AI answer sources with content that fakes its origin or independence." — Relevant Audience