2025-08-15 · Anthropic

Usage Policy Update

agentsresearch

read at source ↗ www.anthropic.com

Usage Policy Update

Source: Anthropic Date: 2025-08-15 URL: https://www.anthropic.com/news/usage-policy-update

Summary

Anthropic updated its Usage Policy effective September 15, 2025, with changes driven by agentic AI risks. Key changes: prohibited malicious cyberattacks and infrastructure compromise (new agentic-specific prohibition); narrowed political content restrictions to target only deceptive/disruptive activities and voter targeting (enabling legitimate policy research); clarified law enforcement language; applied High-Risk Use Case Requirements specifically to consumer-facing outputs rather than B2B interactions. Described as “a living document” responsive to evolving AI risks.

Implications

  • Safety/policy posture / agentic thread. The explicit prohibition of malicious cyberattacks and infrastructure compromise reflects the August 2025 threat intelligence report (ransomware-as-a-service, 17+ extortion targets). The policy is responding in near-real-time to observed abuse patterns.
  • Political content narrowing. Loosening political content restrictions to allow legitimate policy research is a direct response to criticism that Claude was over-restrictive on political topics — the same concern driving the November 2025 political even-handedness evaluation. The narrow carve-out (deceptive/voter targeting remains banned) is the balance.
  • B2B exemption from High-Risk. Removing High-Risk Use Case Requirements from B2B interactions (applying them only to consumer-facing outputs) reduces friction for enterprise API customers building regulated-industry products — a direct enterprise sales enablement change that came in the same period as the financial services and life sciences vertical launches.
  • “Living document” framing. Explicitly positioning the policy as responsive to evolving AI risks rather than a static document is Anthropic’s answer to critics who say AI policy is always behind the capability curve.
  • Watch: whether the cyberattack prohibition actually reduces the attack surface, or attackers simply find different approaches; how the B2B exemption is used in practice; the next policy update trigger.

← all signals