Claude Usage Policy Update 2026 (Nov 12): Developer Checklist
Anthropic announced an update to its Usage Policy on October 8, 2026, and it takes effect on November 12. By Anthropic's own account, some of the changes only clarify existing rules. For developers, the Claude usage policy update has seven changes worth checking against your product: a new section on deceptive activity, a narrower elections section, more detailed weapons and surveillance rules, requirements for high-risk uses, rules for physical hardware, a ban on abuse of the model, and a clarification of supported regions.
This article is based on Anthropic's announcement and reflects the situation as of October 2026. The authoritative text is the Usage Policy and the Supported Regions Policy. The summaries below are not a substitute for the originals.
What exactly changed?
| Change | Key point |
|---|---|
| Deceptive activity | New section, "Do Not Engage in Deceptive Campaigns or Artificial Activity", which merges rules previously scattered across elections, fraud, privacy and misinformation |
| Elections | Renamed "Do Not Undermine Democratic Processes", focused on deceiving voters and interfering with elections |
| Weapons | Now explicitly covers software and components that make a weapon work, including adding weapons to drones and other autonomous vehicles |
| Surveillance and law enforcement | Tracking people without consent is prohibited; Claude may not be used to decide or recommend who to investigate, arrest or prosecute |
| High-risk uses | Rewritten to state which kinds of advice are covered |
| Physical actions | New operator oversight and safe-state requirements for scenarios where Claude is connected to hardware |
| Abuse of the model | New ban on sustained, gratuitous abuse |
What does the deceptive activity section cover?
Anthropic's list includes hiding who is really behind a piece of content, amplifying content with fake accounts or posts, and building tools or infrastructure for influence operations. It applies in both political and commercial contexts. If your product involves bulk posting, comment generation or account management, this is the section to read in full.
Are the election rules tighter or looser?
Both. What is prohibited is deceiving voters and interfering with elections; see Anthropic's original text for the specific examples. Anthropic also removed the blanket ban on personalized voting and campaign targeting. However, targeting that relies on deception or misuse of personal data is still prohibited by other sections.
What changed in surveillance and law enforcement?
The rules are more specific after the rewrite. Tracking someone in real time, or analyzing existing data to track them, is not allowed without consent, and developing or improving surveillance tools is also prohibited. Anthropic says the following are still allowed: consented tracking (for example, fraud monitoring), content moderation, journalism and legal research.
What are the new requirements for high-risk uses and physical hardware?
Products that affect health, legal rights, finances, livelihoods or essential services still need a qualified person who can review and change Claude's recommendations to take part in the decision, and the people affected must be told that AI was used. This requirement is unchanged.
What is new is the rule for physical actions. When Claude is connected to hardware that can take actions on its own and could cause harm, a qualified operator must be able to observe and stop the device, and the device must be in a safe state once Claude is disconnected. Anthropic says this follows its Model Hardware Standard.
Is abusing the model now a violation?
The new clause targets sustained, gratuitous abuse with no clear purpose. Anthropic says it does not cover ordinary frustration, pushback, creative work with dark themes, or model testing and research. In Claude.ai and Claude Code, Claude's ability to end a conversation of sustained abuse remains the main enforcement mechanism.
Did the supported regions rules change?
Anthropic describes this as a clarification. Access is not permitted in these three situations: the person is physically located in an unsupported region; the company is incorporated or headquartered in an unsupported region; or the company is majority-owned or controlled by individuals or entities from those regions. The Supported Regions page has also been updated to explain how this is enforced. Please use the service in line with its terms and local law, and treat the official page as the authority on which regions are supported.
How do developers run through the checks?
Work through the list below in order. Finishing before November 12 is enough:
- Read the announcement, then the relevant sections of the AUP itself, and note the items that apply to your product.
- Does your product generate or distribute public-facing content? Check for hidden publishers, amplification through fake accounts, bulk content farms, and whether you are providing such tools to others.
- Does it involve elections or political content? Make sure there is no impersonation, false voting information or voter suppression.
- Does it have location tracking, behavior tracking or "find a person" features? Make sure there is clear consent. In law enforcement settings, do not let the model decide who is investigated or prosecuted.
- Does it affect health, legal matters, finance, employment or essential services? Make sure a qualified person reviews the output and can change it, and that users are told AI is being used.
- Does it connect to hardware or let Claude take physical actions? Make sure someone can watch in real time and trigger an emergency stop, and that the device goes to a safe state when Claude disconnects.
- Are your company's place of incorporation, headquarters and ownership structure all within supported regions? If you are unsure, consult legal counsel or contact Anthropic.
- Record your findings and the responsible owner in an internal document, with the date.
How do you keep track of changes afterward?
When an official page changes, /en/changes shows line-by-line diffs of changelogs, pricing pages and news pages for Claude and others, with RSS support. If Claude seems unreachable, check /en/down/claude and /en/status/anthropic first to tell a service outage from a problem on your own network. For how to interpret them, see How to read the OpenAI and Claude status pages.
Related reading:
- AI service outages, September-October 2026: OpenAI and Claude incident statistics
- Claude Max 5x vs 20x: what is the difference
- How to find out right away when LLM API prices change
- Troubleshooting Claude Code connection timeouts and errors
