# Anthropic Usage Policy 2026 Update

> Source: https://aiwiki.ai/wiki/anthropic_usage_policy_2026_update
> Updated: 2026-10-10
> Fact-checked: 2026-10-10
> Categories: AI Policy, Anthropic
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Anthropic Usage Policy 2026 Update." aiwiki.ai, 10 Oct 2026. https://aiwiki.ai/wiki/anthropic_usage_policy_2026_update
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

The Anthropic Usage Policy 2026 Update is a published revision of [Anthropic's](https://aiwiki.ai/wiki/anthropic) rules for using [Claude](https://aiwiki.ai/wiki/claude), scheduled to take effect on November 12, 2026. Anthropic describes most changes as clarifications of existing restrictions. [1]

## Earlier policy

The preceding published edition carries an effective date of September 15, 2025. It already separated universal restrictions, requirements for high-risk use, and additional guidelines. Its high-risk requirements included qualified professional review before advice or decisions reached individuals, plus disclosure when AI outputs were presented directly to them. Covered areas included healthcare, legal guidance, finance, employment, housing, academic testing and automatically published journalistic content. [3]

## Deceptive campaigns and elections

The update consolidates restrictions on deceptive political and commercial activity, including concealed sponsorship, fake amplification and influence-operation infrastructure. It removes the blanket prohibition on personalized vote and campaign targeting. Deceptive targeting, misuse of personal data and prohibited surveillance remain barred. The elections section focuses on voter deception and disruption of democratic processes. [1]

## Weapons

Anthropic clarifies that its weapons prohibition includes guidance and control software, weapons components, and arming drones or autonomous vehicles. It describes these changes as consistent with previous enforcement. [1]

## High-risk recommendations and physical actions

The revised policy covers Claude apps, [Claude Code](https://aiwiki.ai/wiki/claude_code), APIs and integrated services. [2]

| Use | Controls or exclusions |
| --- | --- |
| Specified high-risk recommendations, including diagnosis, legal advice and personalized investment advice | Qualified human review, with authority to revise advice or decisions before delivery or implementation, and AI disclosure to affected individuals. Qualification requires relevant training or experience and a license where legally required. [2] |
| Autonomous hardware capable of harmful physical actions | A qualified operator able to observe and stop equipment; independent safety limits; and stopping or holding a safe state after intervention or loss of connection. [2] |
| Related but excluded workflows | General education without individualized advice; internal drafts that are not the delivered recommendation; equipment commands reviewed by a qualified person before execution; and monitoring without controlling equipment. [2] |

Specified wholly favorable care, insurance-claim or public-benefit decisions can bypass prior review where law permits. Partial or conditional approvals do not qualify. Additional legal obligations may still apply. [2]

Surveillance restrictions cover nonconsensual tracking using historical data and building surveillance tools. Criminal investigation, arrest or charging recommendations remain prohibited rather than becoming permitted through human review. [2]

## Extreme cruelty and conversation ending

The new cruelty rule concerns extreme, repeated abuse without a discernible purpose. Anthropic excludes ordinary frustration, pushback, dark creative themes, model testing and research. It says ending rare abusive conversations in Claude.ai and Claude Code remains the primary enforcement mechanism for this conduct. [1]

Anthropic's August 15, 2025 research announcement described conversation ending for Claude Opus 4 and 4.1 in consumer chat interfaces. It framed the feature as experimental [model welfare](https://aiwiki.ai/wiki/model_welfare) work while explicitly acknowledging uncertainty about models' moral status. In that implementation, conversation ending was a last resort after unsuccessful redirection, or a response to a user's request to end a chat. It was not to be used when a user might imminently harm themselves or others. Ending one conversation did not prevent starting another or branching from earlier messages. [4]

## References

1. Anthropic. [2026 Usage Policy Update](https://www.anthropic.com/news/2026-usage-policy-update). October 2026.
2. Anthropic. [Usage Policy](https://www.anthropic.com/legal/aup). Effective November 12, 2026.
3. Anthropic. [Usage Policy, Previous Edition](https://www.anthropic.com/legal/archive/22742366-2ef0-4c7a-a833-6523f10d3944). Effective September 15, 2025.
4. Anthropic. [Claude Opus 4 and 4.1 Can Now End a Rare Subset of Conversations](https://www.anthropic.com/research/end-subset-conversations). August 15, 2025.