Anthropic's updated usage policy for Claude takes effect Nov. 12 and, for the first time, explicitly prohibits using its models to manipulate the sources that search engines and AI systems draw answers from. The clause sits inside a new section titled "Do Not Engage in Deceptive Campaigns or Artificial Activity," released Oct. 8 in a company post listing broader changes covering elections, weapons, surveillance and sustained model abuse.

What does the new clause actually say?

The policy text reads: "Manipulate the sources from which search engines or AI systems draw answers by seeding them with content that misrepresents its origin, authorship, or independence (e.g., networks of sites posing as unaffiliated sources corroborating the same claims)." Anthropic paired that prohibition with bars on fake personas, fake outlets and fake reviews, plus rules against running deceptive-influence tools or selling them as a service. Rules against fake-account networks and fake news sites already existed in earlier versions, the post says, but were scattered across its election, fraud, privacy and disinformation sections.

How does this connect to what Anthropic found in September?

Anthropic's September threat-intelligence report, which the post cites, identified roughly 70 websites posing as independent local news outlets, all linked to a France-based ad agency. The operator used Claude to draft articles in a consistent format, each carrying three to four internal links, and pushed them out through automated publishing. The network spanned around 20 languages and produced at least 8,913 articles, though most drew little engagement from real readers. The report said the articles were "specifically designed to boost their site's authority rankings on search engines" but did not document whether the sites surfaced in AI-generated answers. Anthropic has banned the account tied to the operator.

Does this match Google's approach?

Google's spam policies already label "attempting to manipulate generative AI responses in Google Search" as spam, with a May 15 documentation update confirming that scope. A June spam update rolled out under that policy and a Cornell Tech paper surfaced the same month explaining why enforcement at the source is hard. Google enforces against websites, which can drop them out of rankings; Anthropic enforces against Claude users, and the post warns it "might suspend or terminate access" if it suspects violations. The rules cover API developers and consumers using apps built on Claude.

What's different in the high-risk use list?

The previous version, effective Sept. 15, 2025, never named search engines or AI answers and listed "Media or professional journalistic content" as a high-risk use case covering AI-generated content published externally. The new version names 11 high-risk areas, including legal, medical and financial advice plus credit, housing and employment decisions, but publishing is no longer among them. Anthropic's post says the substantive review and disclosure rules for high-risk recommendations haven't changed, even as the section was rewritten to specify which recommendations fall in scope. The post does not explain why the publishing category disappeared.

What doesn't change?

The rule against submitting "AI-assisted work without proper permission or attribution" carries over, consumer chatbots still must disclose they are AI, and Claude users still can't present outputs as "human-generated" or imply a human is on the other end of a chat.

What should marketers and agencies do by Nov. 12?

The practical audit is narrow. Anyone running a portfolio of sites that look independent of one another, or paying an agency that does, needs to check the funding, staffing and editorial connections across that portfolio. Anyone running paid content pipelines on Claude should recheck the review and disclosure procedures they applied under the older high-risk list against the new list of 11, since legal and financial advice categories tightened in framing even where the requirements stayed constant. Anthropic notes the policy "will continue to be revised as Claude's capabilities and the risks associated with them evolve" and that it updates usage rules annually, so the next reset lands roughly a year out.

Watch for Anthropic's first published enforcement action under the deceptive-campaigns section after Nov. 12.