Anthropic Usage Policy

Anthropic's Usage Policy lists what nobody may use Claude for, such as weapons, malware, child sexual abuse material, fraud, surveillance and election deception, plus extra rules for high-risk uses. The 2026 version takes effect on Nov 12, 2026; violations can be reported to usersafety@anthropic.com.

Anthropic's Usage Policy is the set of rules for what anyone may and may not do with Claude, in the apps, Claude Code and the Claude API alike; the current version, announced on Oct 8, 2026, takes effect on Nov 12, 2026. It bans, among other things, help with weapons, attacks on computer systems and critical infrastructure, child sexual abuse material, fraud, mass surveillance, disinformation and deceptive influence campaigns, and sexually explicit content. Anthropic updates it once a year. Its Safeguards team enforces it with warnings, throttling, suspensions and bans, and Anthropic asks people to report harmful output to usersafety@anthropic.com or with the thumbs-down button (Usage Policy, read Nov 6, 2026).

Quick facts

Current version	Announced Oct 8, 2026, effective Nov 12, 2026
Where to read it	anthropic.com/legal/aup
Applies to	Every Claude product and every company building on Claude
Parts	Universal Usage Standards; High-risk Use Case Requirements; Additional Use Case Guidelines
Report misuse	usersafety@anthropic.com, or the thumbs-down button
Enforcement	Warnings, throttling, limits, suspension, termination
Update cadence	Yearly

What does the Usage Policy ban?

The Universal Usage Standards apply to everyone. On Nov 6, 2026 the policy groups them under these headings (Usage Policy):

Do not violate laws or engage in illegal activity
Do not compromise critical infrastructure
Do not compromise computer or network systems without authorization
Do not develop or design weapons
Do not incite violence or hateful behavior
Do not compromise privacy or identity rights
Do not compromise children's safety
Do not engage in cruel, abusive or psychologically harmful conduct
Do not create or spread disinformation
Do not engage in deceptive campaigns or artificial activity
Do not undermine democratic processes
Do not use Claude for surveillance, censorship, or prohibited law enforcement or criminal justice purposes
Do not engage in fraudulent, abusive or predatory practices
Do not abuse the platform
Do not generate sexually explicit content

What changed in the 2026 update?

The Oct 8, 2026 update made these changes (2026 Usage Policy update):

Deceptive campaigns became their own section, pulling together rules on hidden sponsors, fake accounts and tools for influence operations from the old elections, fraud, privacy and disinformation sections.
Elections became "Do Not Undermine Democratic Processes", and the blanket ban on personalized vote and campaign targeting was dropped.
Weapons now explicitly cover the software and components that make weapons work, and arming drones or other autonomous vehicles.
Surveillance and policing were rewritten: no tracking people without consent, no building surveillance tools, and Claude may not be used to decide or recommend whom to investigate, arrest or charge.
Hardware: when Claude controls equipment that could injure someone, a qualified operator must be able to watch it and stop it, and it must fall back to a safe state if Claude disconnects.
Cruelty toward the model: a new ban on sustained and needless abuse of Claude, with ordinary frustration, dark fiction and research excluded (see Anthropic Safety Research on model welfare).
Supported regions: use is barred for people located in unsupported regions and for companies incorporated there or majority-owned or controlled from there.

High-risk uses

Some uses are allowed only with safeguards. For advice or decisions in eleven areas (legal, medical, financial, credit, insurance, housing, employment, education and credentials, healthcare access, public benefits and legal status), a qualified human must be able to review and change Claude's recommendations, and people must be told AI was involved. Six areas where Claude acts in the physical world carry their own requirements, and companies building agents are responsible for what Claude does through their tools, browsers and connected systems (Usage Policy, read Nov 6, 2026).

Why does Claude refuse some requests?

Refusals have two sources. The first is Claude's training: Claude's Constitution asks Claude to weigh helpfulness against harm, and it sets hard limits (such as serious help toward weapons of mass destruction) that no setting unlocks. A second layer is classifiers: on models covered by the Responsible Scaling Policy, such as Claude Fable 5, separate systems screen chemistry, biology and cybersecurity requests and can block or hand them to a different model. Those classifiers sometimes stop legitimate questions; Anthropic's own tests found they slightly raised refusals of harmless requests (Constitutional Classifiers, Feb 3, 2025).

The constitution also counts needless refusal as a real cost, so a refusal of an ordinary request is a failure by Anthropic's own standard. Rephrasing with the legitimate context (a nurse asking about overdose thresholds, a security team testing its own servers) often helps. Security researchers and life scientists who need more can apply to Anthropic's Cyber Verification Program or Life Sciences Verification Program, which loosen some safeguards for vetted organizations (Anthropic, Oct 6, 2026).

How is the policy enforced?

0:00 Anthropic's Safeguards team watches for violations and may warn, throttle, limit, suspend or terminate access (Usage Policy). In the consumer apps and Claude Code, Claude can also end a conversation that turns persistently abusive. Enforcement reaches companies too: Anthropic estimated in its IPO filing that 5% to 7% of its 2026 revenue came from customers later found to be violating its usage policies (Fortune, Oct 26, 2026). API customers can tag requests with user profiles so an enforcement action hits one end customer instead of the whole organization (API release notes, Oct 20, 2026).

The policy page does not describe an appeal process; it points to Anthropic's Safeguards Support Center for more on enforcement.

How to report misuse

In a chat, use the thumbs-down button under the response to flag it.
Or email usersafety@anthropic.com with what happened (Usage Policy).

Feedback sent this way is kept with the whole conversation for five years (Privacy Center); see Claude Privacy and Data Use.

Frequently asked questions

What won't Claude do? Claude won't help with weapons, malware or attacks on systems you don't own, child sexual abuse material, fraud, surveillance, deceptive influence campaigns or sexually explicit content, under Anthropic's Usage Policy, and its constitution forbids serious help toward weapons of mass destruction under any settings.

Can Claude write sexual content? No. Anthropic's Usage Policy lists generating sexually explicit content among its universal prohibitions for Claude, and no plan or setting unlocks it.

Can I get banned from Claude? Yes. Anthropic can warn, throttle, suspend or terminate a Claude account that breaks the Usage Policy, and it bars use from unsupported regions.

How do I report a harmful Claude response? Use the thumbs-down button under the response, or email usersafety@anthropic.com, as Anthropic's Usage Policy directs.

Can Claude be used for hacking? Claude can help with authorized security work, but the Usage Policy bans compromising systems without authorization; vetted security teams can apply to Anthropic's Cyber Verification Program for fewer restrictions.