The news
Anthropic published an updated usage policy for its Claude models on October 8, 2026, adding rules against deceptive influence campaigns, voter deception and sustained abuse of the model. The company said the changes take effect on November 12, 2026.
The policy is the rulebook every Claude customer agrees to, from consumers to large enterprises building on Anthropic's API. According to Anthropic's announcement, a new section on deceptive activity consolidates rules that bar state media, government propaganda operations and commercial firms from using Claude to run networks of fake accounts or fabricated news sites, or to hide where messages come from.
The elections section has been renamed "Do Not Undermine Democratic Processes." Anthropic said it now focuses on stopping Claude from being used to deceive voters or disrupt elections. The company also said it removed a blanket ban on personalized vote and campaign targeting, narrowing the rule to deceptive or disruptive uses.
Other sections were clarified or rewritten. Weapons restrictions now extend to software and components, including guidance and control software for weapons and arming autonomous vehicles, according to the announcement. The surveillance section bars tracking people without consent, whether from real-time or historical data, and says Claude may not recommend whom to investigate, arrest or charge. Anthropic listed fraud monitoring, content moderation, journalism and legal research as permitted uses.
The update also adds a prohibition on abusive behavior toward the model. Anthropic said it applies only in extreme cases where users repeatedly act cruelly with no discernible purpose, and not to ordinary frustration or model testing. TechCrunch, which reported the change on October 8, noted that Claude was trained in August 2026 to end persistently harmful or abusive conversations; the policy now formalizes that.
Anthropic also clarified its supported-regions rule: use is barred for people physically located in unsupported regions and for entities incorporated, headquartered or majority-owned there.
The numbers
- Policy published
- October 8, 2026
- Effective date
- November 12, 2026
Why CEOs should care
For companies building on Claude, the policy is a contract term, not a press release. Compliance and product teams should map current and planned use cases against the rewritten sections before November 12. Marketing teams running personalized political or advocacy messaging, security teams using Claude for investigations, and any team analyzing location or behavioral data about individuals have the most to review.
The high-risk section matters most for operations. Anthropic says its existing requirement is unchanged: decisions in health, legal, financial, livelihood and essential-services settings require review by a qualified human in the loop. If Claude is part of a lending, hiring, claims or eligibility workflow, CIOs and general counsel should be able to show who that human reviewer is, what they check and how overrides are logged. The new addition is a set of requirements, including safety protocols, for hardware that takes autonomous physical action.
For CISOs and boards, the surveillance and supported-regions language raises vendor-management questions. Ask: do any subsidiaries, contractors or majority owners sit in regions Anthropic does not support? Does any fraud or insider-threat tool cross from monitoring into tracking individuals without consent? Answering those before the effective date reduces the risk of suspended accounts mid-project.
The bigger picture
The rewrite lands weeks before US midterm elections on November 3, 2026, as AI providers face pressure over synthetic propaganda and fake news sites. By narrowing the elections rule to deception and disruption while lifting the blanket targeting ban, Anthropic is drawing a line between legitimate campaign work and manipulation, a distinction regulators and rival providers are also wrestling with. The abuse rule, meanwhile, reflects the company's continued public interest in model welfare, an unusual topic for an enterprise vendor's terms of service.
What’s next
The policy applies from November 12, 2026. Watch for Anthropic's enforcement reporting after the elections, and whether OpenAI, Google and other model providers align their own terms on deceptive campaigns and surveillance.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made by our newsroom, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
Companies in this story
