Anthropic Usage Policy Update August 2025

Anthropic has announced updates to its Usage Policy effective September 15, 2025. These changes refine how Claude is used in agentic contexts, political discourse, law enforcement, and high-risk consumer-facing applications to better align with evolving AI capabilities and regulatory developments.

Cybersecurity and Agentic Use Restrictions

Anthropic is introducing specific prohibitions against malicious computer, network, and infrastructure compromise activities to mitigate risks associated with scaled abuse, malware creation, and cyber attacks. These restrictions are a response to the rapid advancement of agentic capabilities, including tools like Claude Code and Computer Use.

While malicious activities are prohibited, Anthropic continues to support cybersecurity use cases that strengthen security, such as vulnerability discovery conducted with the system owner's consent.

To provide further clarity, Anthropic has published supplementary guidance in its Help Center detailing how the Usage Policy applies to agentic use, including concrete examples of prohibited activities in agentic contexts.

Revised Political Content Guidelines

Anthropic is shifting from a blanket prohibition on all lobbying and campaign content to a more tailored approach. The updated policy specifically prohibits use cases that are:

• Deceptive or disruptive to democratic processes.
• Involved in voter and campaign targeting.

This change allows for legitimate political discourse, policy research, civic education, and political writing, while maintaining restrictions on misleading or invasive activities.

Clarification of Law Enforcement Use

Anthropic has updated the language regarding law enforcement use to be more straightforward and clear. This update does not change the underlying permissions or prohibitions but simplifies the communication of the existing stance.

Restrictions remain in place for the following activities:

• Surveillance
• Tracking
• Profiling
• Biometric monitoring

Support continues for appropriate back-office and analytical use cases that were previously permitted.

High-Risk Consumer-Facing Requirements

Anthropic is clarifying the application of its High-Risk Use Case Requirements. These requirements—which include AI disclosure and human-in-the-loop oversight—apply to use cases with public welfare and social equity implications, such as legal, financial, and employment-related applications.

The policy now explicitly clarifies that these safeguards are required specifically when model outputs are consumer-facing, rather than for business-to-business (B2B) interactions.

Policy Evolution and Governance

Anthropic treats its Usage Policy as a living document. The company stated it will continue to evaluate these policies through collaboration with external policymakers, civil society, and subject matter experts to adapt to evolving AI risks.