Anthropic has announced a notable update to its Usage Policy, which will go into effect one month from now on September 15. The changes reflect the growing capabilities of its products, namely agentic capabilities. The goal of the update is to give users more clarity on the rules for Claude as the technology becomes more powerful. Anthropic even describes the Usage Policy as a “living document” that will continue to change as AI risks evolve.
The update adds a new section to the policy specifically about agentic capabilities. With new tools such as Claude Code and Computer Use, there are new risks which justify the update. The new Usage Policy now prohibits activities including malicious computer, network, and infrastructure compromise activities. With that said, the company is continuing to support beneficial cybersecurity use cases, such as discovering vulnerabilities, if you have the proper consent.
These updates follow a previously published threat intelligence report which flagged risks like malware creation and cyber attacks. As a supplement, Anthropic has also released a separate Help Center article that gives detailed examples of prohibited agentic use.
Aside from addressing the dangers of agentic AI, the company is tailoring its restrictions on political content. The company has eased up on the matter, removing blanket prohibitions on all political content in favor of targeted restrictions on deceptive or disruptive uses. The policy also clarifies that high-risk use case requirements, such as human oversight for legal, financial, and employment applications, apply only to consumer-facing outputs.
The frontier AI company said the changes have been made in response to user feedback, which said that the old rules were too broad and limited legitimate research and writing.
With this update, Anthropic’s policies attempt to strike a careful balance between enabling powerful new tools and preventing their misuse. The company said it’s working with external policymakers, subject matter experts, and civil society to evaluate its policies on an ongoing basis.