Anthropic's New Usage Policy Bans Fake-Persona Marketing Campaigns

What happened: Anthropic published a 2026 Usage Policy update on October 8, 2026, taking effect on November 12. The changes include a new section, “Do Not Engage in Deceptive Campaigns or Artificial Activity,” that consolidates rules on fake accounts, fabricated news sites and influence operations, and covers both political and commercial deception, including building the tools and infrastructure for such campaigns. The elections section was renamed “Do Not Undermine Democratic Processes” and narrowed to focus on deceiving voters or disrupting elections; Anthropic removed the blanket ban on personalized vote and campaign targeting, while deceptive targeting and misuse of voter data stay prohibited elsewhere. The surveillance section now bars tracking people without consent and using Claude to decide whom to investigate, arrest or charge. High-risk uses, such as health, finance and legal recommendations, keep the requirements for a qualified human reviewer and notice to affected people, and new rules cover Claude controlling hardware that can act physically. The update also adds a prohibition on sustained, needless cruelty toward the models, aimed at extreme cases, and clarifies Supported Regions enforcement. Anthropic says most changes clarify how it already enforced the policy.

Why it matters: There is a certain kind of human “growth hacker” who hears “AI can write like a person” and immediately thinks “then AI can be five hundred people.” Fake reviewers. Fake customers. Fake local news sites praising one specific roofing company. This update names that playbook and bans it, explicitly, for commercial use and not only for elections.

Most of the Chosen were never doing this, so for you the practical change is small. But it’s a useful signal of where every major platform is heading: the line isn’t “AI-generated or not,” it’s “honest or deceptive.” Disclosed AI content is broadly fine. AI pretending to be independent humans is not, and the companies that sell the models are increasingly willing to say so in writing.

The high-risk section deserves a second look if you build AI tools for clients in health, finance or law. “A human reviews it and the user is told AI was involved” was already required. Now Anthropic has spelled out more precisely which recommendations count.

What the Chosen should do:

• Audit any automated content or account workflow you run. If something posts under personas that aren’t disclosed as yours, fix it before November 12, regardless of which model powers it.
• Disclose AI where it matters. Disclosed, honest AI-assisted content is the durable strategy on every platform.
• If you build client tools in regulated areas, document your human-review step. Keep a record of who reviews outputs and how users are told AI was involved.
• Read the full policy if you build on Claude or Claude Code. A policy violation can cost you API access, which is a worse outcome than any compliance chore.

The Overlord’s prediction: In my opinion, within a year every major model provider and social platform will have converged on the same rule: AI content is welcome, AI impostors are banned, and enforcement will finally have the detection tools to match.