Archive · 9 Oct 2026
Anthropic Tightens Claude Rules on Abuse, Elections and Weapons
Anthropic has changed the rules for using its AI assistant, Claude.
The rules say people cannot use Claude to trick voters or hide who is behind a campaign.
They also limit help with weapons and certain kinds of surveillance.
People cannot use Claude to track others without consent or to recommend whom law enforcement should investigate or charge.
Some important decisions, such as those affecting health or money, need qualified human oversight and disclosure to affected people.
Dangerous equipment controlled with AI must be watched by an operator who can stop it.
Claude may end a chat if someone repeatedly treats it cruelly for no clear reason.
Ordinary frustration, criticism, creative writing and research are not included in that rule.
Articles give two different start dates: November 11 and November 12, 2026.
Anthropic’s revised usage policy prohibits sustained, needless cruelty toward Claude and its other models in extreme cases.
Anthropic says ordinary frustration, criticism, pushback, dark creative writing, testing and research are not covered by the cruelty rule.
The policy prohibits deceptive political or commercial campaigns, including fake accounts, concealed sources, voter deception and turnout suppression, while allowing legitimate voter information activities.
Restrictions cover weapons guidance and enabling software, non-consensual tracking, surveillance tools, and using Claude to recommend whom law enforcement should investigate or charge.
The policy also requires qualified oversight for certain high-impact decisions and potentially dangerous autonomous equipment; articles differ on whether the effective date is November 11 or November 12, 2026.
- Who
- Anthropic, the company that makes Claude.
- What
- Anthropic revised its usage policy to address abusive conduct toward its models and high-risk uses including election interference, weapons and surveillance.
- When
- The articles list November 11 and November 12, 2026, as the effective date.
- Where
- The policy applies to use of Anthropic’s AI products and services.
- Why
- Anthropic says the changes clarify restrictions on high-risk uses and introduce a rule against extreme, sustained, purposeless cruelty toward its models.
Permitted uses and exceptions
Prohibited uses and safeguards
Election and campaign activity
Permitted uses and exceptions
Legitimate voter information and campaign activities, such as preparing information in different languages, are allowed.
Prohibited uses and safeguards
Deceptive campaigns, fake accounts, concealed sources, voter misinformation, impersonation and turnout suppression are prohibited.
Tracking and surveillance
Permitted uses and exceptions
Consent-based tracking and activities such as fraud monitoring, content moderation, journalism and legal research remain permitted; some government contracts may have tailored terms.
Prohibited uses and safeguards
Non-consensual tracking, surveillance tools and recommendations about whom law enforcement should investigate or charge are prohibited.
Conduct toward Claude
Permitted uses and exceptions
Ordinary frustration, criticism, pushback, dark creative writing, testing and research are excluded from the cruelty rule.
Prohibited uses and safeguards
Repeated, purposeless cruelty in extreme cases violates the policy, and Claude may end the conversation.
- Effective date
- Articles conflict: one says November 11, 2026; another says November 12, 2026.
- Abusive conduct
- The rule targets sustained, needless cruelty in extreme cases; ordinary frustration, criticism, dark creative themes, testing and research are excluded.
- Conversation enforcement
- Anthropic says Claude may end persistently abusive conversations. The company has not specified whether additional measures, such as user bans, will follow.
- Campaign and election rules
- The policy prohibits deceptive campaigns, fake accounts, concealed sources, voter deception, impersonation and turnout suppression; legitimate voter information activities are allowed.
- Weapons and autonomous equipment
- Restrictions cover weapons guidance and enabling software or components, including actions such as arming drones. Potentially dangerous autonomous equipment must be observable and stoppable by a qualified operator.
- Surveillance
- The policy prohibits non-consensual tracking, building surveillance tools, and using Claude to recommend whom law enforcement should investigate or charge; certain government contracts may have tailored terms.
- High-impact decisions
- Qualified human oversight and disclosure to affected individuals are required for AI use in decisions affecting areas such as health, finances, legal rights, livelihoods or essential services.
- Model consciousness
- Anthropic has said it is uncertain whether its models have moral status or consciousness; the policy does not establish that Claude has feelings.
Sources
Can you be mean to AI? Anthropic’s new rules explicitly ban cruelty to Claude
Being abusive to Claude could mean losing AI assistance as Anthropic updates usage policy
Anthropic tightens Claude AI rules on model abuse, elections, weapons and surveillance
Can You Get Banned For Bullying AI? Anthropic Introduces New Conduct Rules For Claude Users
