Anthropic published an update to its Usage Policy on August 15, 2025, revising how its Claude models should and should not be used as their capabilities and usage patterns evolve. The changes become effective on September 15, 2025.
Why the change?
The company says agentic capabilities expanded significantly over the past year: Anthropic released its own agentic tools, including Claude Code and Computer Use, and its models power many leading coding agents. These advances introduce new risks such as potential scaled abuse, malware creation, and cyber attacks — risks the company previously documented in its threat‑intelligence report “Detecting and Countering Malicious Uses of Claude: March 2025.”
Cybersecurity and agentic use: explicit prohibitions
The updated Usage Policy adds a section that explicitly prohibits malicious computer, network, and infrastructure compromise activities. At the same time, Anthropic continues to permit and support legitimate cybersecurity use cases, for example vulnerability discovery performed with the system owner’s consent.
Anthropic also published a Help Center article explaining how the Usage Policy applies more broadly to agentic use. That supplementary guidance provides concrete examples of prohibited agentic activities and is not intended to replace the formal Usage Policy.
Narrowing broad political content restrictions
Historically, Anthropic’s Usage Policy contained wide prohibitions on lobbying and campaign content due to the unknown risks of AI‑generated content affecting democratic processes. User feedback indicated that those blanket bans also constrained legitimate uses of Claude for policy research, civic education, and political writing.
The revised policy narrows that approach: it now specifically forbids use cases that are deceptive or disruptive to democratic processes and any voter or campaign targeting. This change is intended to allow legitimate political discourse and research while prohibiting misleading or invasive activity.
Law‑enforcement language clarified without substantive change
Earlier language around law‑enforcement use included multiple exceptions for back‑office tools and analytic applications, which sometimes made permitted use cases hard to parse. The update rephrases these sections to be clearer and more straightforward, but Anthropic says this does not change what is allowed or prohibited in practice. The company continues to restrict surveillance, tracking, profiling, and biometric monitoring while supporting appropriate back‑office and analytic uses that were already permitted.
Requirements for high‑risk consumer‑facing use cases
Anthropic’s High‑Risk Use Case Requirements cover applications with public‑welfare and social‑equity implications — such as legal, financial, and employment‑related uses of Claude — and require additional safeguards like human‑in‑the‑loop oversight and AI disclosure.
As Claude’s enterprise usage has expanded, Anthropic clarifies that these requirements apply specifically when model outputs are consumer‑facing, not to business‑to‑business interactions.
Looking ahead
Anthropic frames the Usage Policy as a living document that will evolve alongside AI risks. The company says it will continue to review and update policies in consultation with internal teams and external stakeholders, including policymakers, subject‑matter experts, and civil society.
Key dates and references
- Update published: August 15, 2025
- Effective date: September 15, 2025
- Referenced threat report: “Detecting and Countering Malicious Uses of Claude: March 2025”
- Mentioned products: Claude, Claude Code, Computer Use
