Anthropic updates Claude policy for cyber, political and high-risk uses

Anthropic updated its Usage Policy for Claude, adding rules for malicious cyber activity, narrowing broad restrictions on political content and clarifying how existing requirements apply to law enforcement and high-risk consumer-facing uses.

The company said the changes were intended to provide more detail and clarity in response to user feedback, product changes, regulatory developments and its enforcement priorities. The revised policy was scheduled to take effect on September 15, 2025.

Cybersecurity and agentic use

Anthropic added a section describing prohibited activities involving malicious compromise of computers, networks and infrastructure. The company cited risks including abuse at scale, malware creation and cyberattacks as agentic capabilities have advanced and products such as Claude Code and Computer Use have been released.

The revised rules continue to support cybersecurity work intended to strengthen systems. Anthropic gave vulnerability discovery conducted with the system owner’s consent as an example of supported activity.

Anthropic also published supplementary Help Center guidance explaining how its Usage Policy applies to agentic uses. According to the company, that article contains concrete examples of prohibited activities involving agents but does not replace or supersede the Usage Policy.

Political content restrictions

The update revises Anthropic’s previous broad prohibitions on lobbying and campaign content. The company said users had reported that the blanket restrictions also limited uses of Claude for policy research, civic education and political writing.

Under the revised approach, Anthropic said it is specifically prohibiting uses that are deceptive or disruptive to democratic processes, as well as voter and campaign targeting. The company said the narrower rules permit legitimate political discourse and research while barring misleading or invasive activity.

Law enforcement language

Anthropic rewrote its policy language concerning law enforcement uses, saying earlier wording and its exceptions for back-office tools and analytical applications had sometimes made the boundaries difficult to understand.

The company said the revision changes the wording, not the activities that are permitted or prohibited. Restrictions remain in place for surveillance, tracking, profiling and biometric monitoring. Anthropic also continues to permit the appropriate back-office and analytical applications that were allowed under the previous policy.

High-risk consumer-facing uses

Anthropic clarified the scope of its High-Risk Use Case Requirements, which cover applications with public welfare and social equity implications. The company identified legal, financial and employment-related uses of Claude among the areas subject to these requirements.

Those applications require safeguards including human oversight and disclosure that AI is being used. Anthropic said the revised language makes clear that these requirements apply when model outputs face consumers, rather than to business-to-business interactions.

Anthropic described its Usage Policy as a document it expects to update as AI risks change. The company said it would continue evaluating its policies internally and with external policymakers, subject matter experts and civil society.

Source: Anthropic’s “Usage policy update,” published August 15, 2025.