Anthropic has updated its Usage Policy for Claude, introducing explicit restrictions on prolonged abusive behaviour toward its AI models and tightening rules covering election interference, deceptive campaigns, weapons development and surveillance. Announced on October 8, 2026, the revised policy responds to new patterns of misuse as Claude becomes capable of handling longer, more independent tasks. The company said the changes clarify how its existing safety principles apply to increasingly capable AI systems, while introducing more precise requirements for certain high-risk applications. The updated rules are scheduled to take effect on November 12, 2026.
The most notable change is a new prohibition against sustained and needless abusive or cruel behaviour toward Claude, although Anthropic says it applies only in extreme cases. The company has also consolidated restrictions on deceptive political and commercial campaigns, clarified its rules against election manipulation and expanded language covering weapons-related software and surveillance systems. According to Anthropic’s official announcement, most of the revisions are intended to make existing restrictions clearer rather than introduce entirely new enforcement standards.
Anthropic Adds a Rule Against Abusing Claude
Anthropic’s updated policy explicitly prohibits sustained and needless abusive or cruel behaviour toward its AI models. The restriction is designed for situations in which users repeatedly engage in cruelty toward Claude without a discernible purpose.
The company has emphasised that the rule is narrow. Ordinary frustration, criticism of an incorrect response, challenging the model’s reasoning, dark creative writing and legitimate model testing or research are not covered by the prohibition.
Claude has already been able to end certain conversations involving persistently harmful or abusive interactions on Claude.ai and Claude Code. Anthropic said conversation termination will remain the primary enforcement mechanism under the revised policy.
The change follows the company’s research into AI welfare and model behaviour. Anthropic has previously explored whether AI systems exhibit preferences or apparent distress during interactions, while acknowledging uncertainty about whether such behaviour establishes that a model has any form of moral status.
The new policy does not establish that Claude is conscious or has feelings. Instead, it sets a boundary around extreme interactions that Anthropic considers unnecessary and abusive.
For users, the practical distinction is between legitimate criticism and repeatedly directing cruelty at the model for its own sake. The policy does not prevent people from questioning Claude’s answers, testing its limits or reporting poor performance.
Election Interference and Deceptive Campaigns Face Clearer Restrictions
Another major change concerns the use of Claude to manipulate political discussion, mislead voters or undermine democratic processes.
Anthropic has introduced a consolidated section titled “Do Not Engage in Deceptive Campaigns or Artificial Activity.” It brings together restrictions that were previously spread across several sections of the policy.
The rules prohibit using Claude to obscure the origin of messages, amplify content through fake accounts or posts, or build tools and infrastructure for deceptive influence operations. The restrictions apply to political and commercial activity.
Anthropic said it had observed state media outlets, government propaganda offices and commercial firms using Claude to operate networks of fake accounts and fabricated news websites.
The company has also renamed and refined its election-related restrictions under the heading “Do Not Undermine Democratic Processes.”
Examples of prohibited activity include:
- Spreading false information about candidates or voting procedures.
- Impersonating election officials or political candidates.
- Attempting to suppress voter turnout through deception.
- Interfering with the integrity of election processes.
- Creating deceptive campaigns that disguise who is responsible for political messages.
These restrictions are intended to prevent AI tools from being used to manipulate voters or manufacture the appearance of genuine public support.
Anthropic removes a blanket restriction on personalised political targeting
The revised policy also removes a blanket prohibition on personalised vote and campaign targeting.
Anthropic said the earlier restriction could unintentionally cover legitimate civic activities, such as nonprofits producing voting information in different languages or election officials contacting voters about ballot corrections.
The removal does not mean all forms of political targeting are now permitted. Deceptive campaigns, misuse of personal data and prohibited surveillance remain restricted under other parts of the policy.
The revised approach focuses more directly on harmful conduct, rather than treating every form of personalised political communication as inherently prohibited.
Weapons Restrictions Now Explicitly Cover Software and Components
Anthropic has clarified that its prohibition on using Claude to develop weapons extends to the software and components that make weapons operate.
The company said it had observed attempts to use its models to develop guidance and control software for weapons. The revised policy makes the restrictions more explicit for activities involving autonomous systems.
Prohibited uses include developing software or components for weapons and arming drones or other autonomous vehicles.
The policy also strengthens restrictions involving biological and chemical weapons, including activities intended to increase the lethality or transmissibility of harmful biological agents.
These changes reflect the growing ability of AI systems to assist with technical work beyond generating general information. Software that controls a physical system can be as important to its operation as its mechanical components.
Anthropic’s position is that its models should not be used to facilitate prohibited weapons development, even when the requested assistance focuses on software rather than the physical weapon itself.
Surveillance and Law Enforcement Rules Become More Precise
The revised policy also clarifies how Claude may be used in surveillance and criminal justice applications.
Anthropic prohibits tracking individuals without their consent, whether the tracking takes place in real time or relies on previously collected data. It also prohibits using Claude to build or improve tools designed for surveillance.
In addition, Claude cannot be used to decide or recommend which people should be investigated, arrested or charged in law enforcement or criminal justice processes.
These restrictions address concerns about AI systems being used to monitor individuals, identify targets or influence decisions that can significantly affect people’s rights.
The policy does not prohibit every use of tracking or monitoring technology. Anthropic identifies permitted activities such as fraud monitoring with appropriate consent, content moderation, journalism and legal research, subject to the applicable restrictions.
The distinction is between legitimate, authorised activities and systems that enable unauthorised monitoring or automated decisions about individuals in criminal justice settings.
New Safeguards for High-Risk Applications and Autonomous Hardware
Anthropic has also clarified its requirements for applications that could affect people’s health, finances, legal rights, livelihoods or access to essential services.
For covered high-risk use cases, a qualified human must be able to review Claude’s recommendations and change them when necessary. Affected individuals must also be informed when AI has been used.
Anthropic said these requirements were already part of its policy. The revision makes their scope easier to understand by identifying the types of decisions that fall within the rules.
The updated policy introduces additional requirements when Claude is connected to hardware capable of taking autonomous physical actions that could cause injury.
In such cases, a qualified operator must be able to monitor the equipment and stop it when necessary. The system must also be capable of maintaining a safe state if Claude becomes disconnected.
These safeguards are intended to address the risks associated with AI systems that can influence physical environments rather than simply produce digital outputs.
Key Changes at a Glance
| Policy area | Main change |
|---|---|
| Model abuse | Explicitly prohibits sustained, needless abusive or cruel behaviour |
| Elections | Clarifies restrictions on deceiving voters and disrupting democratic processes |
| Deceptive campaigns | Consolidates rules against fake accounts, fabricated websites and disguised influence operations |
| Political targeting | Removes the blanket ban while retaining restrictions on deception and misuse of personal data |
| Weapons | Explicitly covers weapons-related software, components and arming autonomous vehicles |
| Surveillance | Clarifies restrictions on non-consensual tracking and surveillance tools |
| Criminal justice | Prohibits using Claude to decide or recommend who should be investigated, arrested or charged |
| High-risk applications | Clarifies human-review and disclosure requirements |
| Autonomous hardware | Requires qualified oversight and the ability to stop potentially dangerous equipment |
The updated policy takes effect on November 12, 2026. Users and businesses relying on Claude should review the full policy to understand how the changes apply to their particular activities.
Why Anthropic Is Updating Its Rules Now
Anthropic reviews its Usage Policy as its models become more capable and the company identifies new forms of misuse.
The latest update follows the expansion of Claude’s ability to handle longer workflows, interact with external tools and perform more independent work. Those capabilities can improve productivity, but they can also make AI assistance more consequential when used for deception, surveillance or dangerous technical applications.
The revisions also reflect findings described in Anthropic’s threat intelligence reporting, which has documented attempts to use AI in influence operations, cyberattacks and surveillance.
For businesses, the changes highlight the importance of reviewing how AI systems are deployed, particularly when they are connected to sensitive data, operational software or physical equipment.
The policy does not eliminate misuse by itself. Its effectiveness will depend on enforcement, technical safeguards, monitoring and users’ compliance with the rules.
The Bigger Picture
Anthropic’s updated Usage Policy reflects the challenges of governing AI systems that can do more than answer questions. As models gain the ability to plan, use tools and support complex technical work, restrictions must address not only the content they generate but also the activities they help users carry out.
The changes combine stronger clarity around prohibited behaviour with more targeted rules for legitimate uses. The removal of the blanket restriction on personalised political targeting illustrates this approach: the objective is to prevent deception and harmful manipulation without unnecessarily blocking lawful civic communication. Ultimately, the policy’s effectiveness will depend on whether the rules can distinguish harmful activity from legitimate work in practice.
Looking Ahead
The revised policy is scheduled to take effect on November 12, 2026. Anthropic is expected to continue refining its safeguards as it observes how customers use Claude and identifies emerging risks. Businesses using the platform will need to check whether their workflows comply with the clarified requirements, particularly in political communication, high-risk decision-making, surveillance and autonomous hardware.
Over the longer term, the update highlights a broader industry challenge: enabling powerful AI tools while limiting their use for manipulation, abuse and dangerous activities. Clear rules are an important part of that effort, but they must be supported by reliable technical safeguards and consistent enforcement. For Claude users, the central takeaway is that ordinary criticism and legitimate testing remain permitted, while sustained needless abuse and the specified harmful uses are explicitly prohibited.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
