Anthropic published its 2026 Usage Policy update on October 8, 2026, with the full text on its Usage Policy page. “The updated policy takes effect on November 12.” Anthropic’s own summary of the reason: “In the year since our last refresh, Claude has taken on longer, more independent work.” Most of the edits, it says, “are intended to clarify existing rules.”
This page lists what changed, section by section, from Anthropic’s post. Every quotation is from it.
On this page · 6 sectionsOpen
Anthropic published its 2026 Usage Policy update on October 8, 2026. It takes effect on November 12, and Anthropic says most of the changes clarify existing rules for a Claude that now does longer, more independent work.
A new section, Do Not Engage in Deceptive Campaigns or Artificial Activity, gathers rules that were scattered across four sections and covers fake accounts, hidden authorship and the tools that run influence operations, political or commercial.
The elections section is renamed Do Not Undermine Democratic Processes and narrowed to deceiving voters, while the blanket ban on personalized vote and campaign targeting is removed because it blocked legitimate civic work.
High-risk uses in health, legal rights, finance and essential services still need a qualified human who can change Claude’s recommendation and a notice that AI was used; the section now lists which recommendations are covered.
New controls for Claude driving hardware that moves: a qualified operator must be able to watch and stop it, and the equipment must hold a safe state if Claude disconnects. A narrow new rule also bars sustained, needless abuse of the models.
§ 01What changed, section by section
| Area | What changed | New rule or clarification |
|---|---|---|
| Deception | Rules from the elections, fraud, privacy and disinformation sections merged into Do Not Engage in Deceptive Campaigns or Artificial Activity | Consolidation, now explicitly political or commercial |
| Elections | Renamed Do Not Undermine Democratic Processes; blanket ban on personalized vote and campaign targeting removed | Narrower rule |
| Weapons | Prohibitions now name guidance and control software, components, and arming drones and other autonomous vehicles | Clarification of past enforcement |
| Surveillance and law enforcement | Tracking without consent banned whether live or from stored data; no deciding or recommending who to investigate, arrest or charge; no building surveillance tools | Clarification of past enforcement |
| High-risk use cases | Rewritten to list which recommendations need a human in the loop and an AI notice | Same requirements, clearer scope |
| Physical actions | Hardware that takes autonomous physical actions needs an operator who can watch and stop it, and a safe state on disconnect | New rule |
| Abuse of models | Sustained, needless abuse or cruelty toward the models prohibited | New rule, extreme cases only |
| Supported regions | Clarifies the ban covers people located in, entities headquartered in, and entities majority-owned from unsupported regions | Clarification |
§ 02Deception gets its own section
The new section is the direct answer to the September threat report. Anthropic writes that it has “seen examples of state media outlets, government propaganda offices, and commercial firms using Claude to run networks of fake accounts and fabricated news sites.” The rules existed but were spread across four sections. The new one “covers efforts to obscure who is behind a message or amplify content through fake accounts or posts, along with building the tools and infrastructure for running influence operation campaigns.” The same day, OpenAI published its own report on two false-front operations.
The elections section moves the other way. Anthropic dropped the blanket ban on personalized vote and campaign targeting because it “covered legitimate civic work, such as nonprofits using Claude to write information for voters in other languages, or election officials sending ballot cure notices.”
§ 03High-risk work and machines that move
The high-risk requirements did not change, but they now come with a list. Where Claude’s output can affect “someone’s health, legal rights, finances, livelihood, or access to essential services,” Anthropic requires “a qualified human in the loop (that is, someone who has the authority to review and, if necessary, change Claude’s recommendations), and for the individual affected to be told that AI was used.”
The physical rule follows Anthropic’s Model Hardware Standard for agents that operate lab instruments:
| Requirement | Anthropic’s words |
|---|---|
| Human oversight | “A qualified operator must be able to observe the equipment and stop it if needed” |
| Fail safe | “the equipment must also be able to hold a safe state if Claude is disconnected” |
The abuse rule is narrow by design. Anthropic says it “does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research,” and that Claude’s ability to end abusive conversations “will remain the primary enforcement mechanism.”
§ 04What this means if your AI employees run on Claude
Our conflict, declared: we build CellCog, and every tier of our AI employees runs on Claude Opus 5.5, so this policy applies to the work they do for you. Two lines matter most. Work that affects someone’s health, money or legal rights needs a person with the authority to change the outcome, and the person affected has to be told AI was involved. On CellCog, consequential actions wait for your approval and every employee says it is AI, which helps; the qualified human is still you or someone you name.
The second is the deception section. An AI employee that writes and publishes for a business must do it under the business’s own name, never through fake accounts or invented authors, and that has been how ours work from day one.
§ 05What we are watching for
- November 12. The policy takes effect; any enforcement notes Anthropic publishes after that date.
- The high-risk list itself. The rewritten section on the Usage Policy page is the text that settles whether a given use needs a human in the loop.
- OpenAI’s usage policies. Whether OpenAI moves the same way on deception and robotics.
§ 06Sources
- Anthropic, 2026 Usage Policy update, October 8, 2026 (published 17:00 UTC); every quotation above is from this post.
- Anthropic, Usage Policy, the full text, effective November 12, 2026.
- Anthropic, Model Hardware Standard research preview.
Q1What counts as a deceptive campaign under the new section?
Anthropic says it covers efforts to obscure who is behind a message or to amplify content through fake accounts or posts, plus building the tools and infrastructure that run influence operations, whether the aim is political or commercial.
Q2Can Claude still be used for voter outreach?
Yes, for legitimate work. Anthropic removed the blanket ban on personalized vote and campaign targeting, naming nonprofits writing voter information in other languages and election officials sending ballot cure notices. Deceptive targeting and misuse of voters’ personal data stay prohibited under other sections.
Q3What are the high-risk requirements?
Where Claude’s output can affect someone’s health, legal rights, finances, livelihood or access to essential services, Anthropic requires a qualified human with authority to review and change the recommendation, and that the affected person be told AI was used.
Q4What does the abuse rule mean for ordinary users?
Little. Anthropic says it applies only to extreme, repeated, purposeless cruelty toward the models, not to frustration, pushback, dark creative themes or testing, and that Claude ending such conversations stays the main enforcement.
Q5Does this apply to work done through CellCog?
Yes. Every CellCog tier runs Claude Opus 5.5, so Anthropic’s Usage Policy applies to what our AI employees do for you, alongside our own terms.