Anthropic's 2026 usage policy: what changes on 12 November

Anthropic rewrites its rules for Claude once a year, and the 2026 version is out. Anthropic published it on 8 October; the new Usage Policy takes effect on 12 November 2026. Most of it clarifies rules that already applied, with new examples for a Claude that now works longer and more on its own. A few points are genuinely new.

The changes in short

• Deceptive campaigns get their own section. Rules on fake accounts, fabricated news sites and influence operations used to be spread across the elections, fraud, privacy and disinformation parts. The new section "Do Not Engage in Deceptive Campaigns or Artificial Activity" covers deception of any kind, political or commercial, including hiding who is behind a message and building the tools and infrastructure for such campaigns.
• Elections, narrower and clearer. The section is now called "Do Not Undermine Democratic Processes" and targets deceiving voters and disrupting elections. The blanket ban on personalized vote and campaign targeting is gone, because it also caught legitimate civic work such as voter information in other languages.
• Weapons include the software. Guidance and control software for weapons and arming drones or other autonomous vehicles are now named explicitly. Anthropic says this matches how it already enforced the old policy.
• Surveillance, stated precisely. The policy prohibits non-consensual surveillance, certain criminal-justice recommendations and tools designed for these prohibited uses. It allows specified uses such as consensual fraud monitoring, content moderation, journalism and legal research; the full policy defines the boundaries.
• High-risk uses, explained better. For advice or decisions in the covered high-risk uses, a qualified human must meaningfully review the output before it is used and be able to correct it. People affected must be told about the AI use. The policy lists the covered areas, including health, legal rights and finances, and its exceptions.
• Robots and other hardware. When Claude drives equipment that takes physical actions and could injure someone, a qualified operator must be able to watch and stop it. Operating limits must be enforced independently of the model, and the equipment must hold a safe state if Claude is disconnected.
• Abusing the model. Sustained and needless cruelty toward Claude is now prohibited. Anthropic stresses that this is for extreme cases only, not for frustration, pushback, dark creative themes or testing.
• Supported regions. The separate regions policy now says plainly that Claude may not be used by people located in an unsupported region, by companies incorporated or headquartered there, or by companies majority-owned or controlled from there.

Our take

For people who code with agents, two parts deserve a careful read. The deceptive-campaigns section names the tooling, not just the posting: an agent that builds sockpuppet account generators or astroturfing pipelines is now plainly out of bounds, whoever clicks publish. And the hardware rules arrive just as agents start to drive real devices. Everything else is mostly a clearer map of rules that were already there, which is a good thing in a policy.

In AyCode

AyCode does not change these rules. A Claude chat in AyCode runs Claude Code as it is signed in on your computer, or with an API key, so Anthropic's usage policy applies there exactly as in the terminal. How AyCode runs Claude Code, Codex and the other agents side by side is in the docs on providers.