2026 Usage Policy: what changed, what to tell Claude
Anthropic's new Usage Policy takes effect November 12, 2026, and tightens how Claude can be used for elections, weapons, surveillance, and high-risk decisions. This lesson shows what to say to Claude so your prompts and products stay inside the new rules.
Anthropic published a new Usage Policy on October 8, 2026. It takes effect November 12, 2026, and changes what Claude will and won't do for you — know these changes now so your prompts and products stay compliant.
Most changes clarify existing rules rather than add new ones. Anthropic rewrote sections to match how Claude is used today: longer, more independent tasks, autonomous control of physical hardware, and new misuse patterns in influence operations, weapons development, and surveillance.
Six things changed: a new section on deceptive campaigns, a narrower elections section, clearer weapons prohibitions, more precise surveillance language, clearer high-risk use case rules (now covering autonomous physical hardware), and a new ban on sustained, needless abuse directed at Claude.
Deceptive activity now has one home. The new section, "Do Not Engage in Deceptive Campaigns or Artificial Activity," consolidates rules that used to be scattered across elections, fraud, privacy, and disinformation sections. It bans hiding who's behind a message, running fake-account networks, and building infrastructure for influence operations — political or commercial.
The elections section was renamed "Do Not Undermine Democratic Processes" and narrowed to target deception and disruption specifically: false voting information, impersonating candidates or officials, suppressing turnout. Anthropic also removed its old blanket ban on personalized vote and campaign targeting — that rule had also blocked legitimate work, like a nonprofit translating voter information or an election office sending ballot cure notices. Deceptive targeting is still banned, just under the deceptive-campaigns and privacy sections instead.
Weapons rules now explicitly cover software. Claude was already banned from being used to build weapons; the update clarifies this includes guidance and control software, and actions like arming a drone or other autonomous vehicle.
Surveillance rules got more precise, not stricter: tracking people without consent is banned whether it happens in real time or on stored data later, and Claude can't be used to decide or recommend who to investigate, arrest, or charge. Consented tracking (like fraud monitoring), content moderation, journalism, and legal research all remain fine.
High-risk use cases — anything touching someone's health, legal rights, finances, livelihood, or access to essential services — still require a qualified human who can review and change Claude's recommendation, plus telling the affected person AI was used. That rule isn't new, but the section now directly answers "does my use case need this?" It also adds a requirement for Claude connected to physical hardware that could cause injury: a qualified operator must be able to watch the equipment and stop it, and the equipment must fail safe if Claude disconnects.
Separately, the policy now bans sustained, needless cruelty toward Claude itself — not normal frustration or pushback, just extreme, repeated abuse with no purpose.
Before shipping anything high-risk or hardware-connected after November 12, 2026: tell Claude explicitly what your human-in-the-loop setup is, make sure the affected person is told AI was involved, and for physical hardware confirm a human can observe and stop the equipment and that it fails safe on disconnect. For anything touching elections or influence, the test is simple: does this deceive or disrupt, or is it transparent, factual work?