Anthropic publishes a 2026 update to its usage policy

Anthropic published a revised version of its Usage Policy, due to take effect 12 November 2026. The update consolidates rules against fake accounts, fabricated news sites and influence operations into a single “Do Not Engage in Deceptive Campaigns or Artificial Activity” section, and narrows the policy’s previous elections section — renamed “Do Not Undermine Democratic Processes” — to focus specifically on voter deception and election disruption, dropping a blanket ban on personalised political targeting that Anthropic said had also constrained legitimate civic organising work.

The revision extends its weapons prohibitions to explicitly cover guidance software, control systems and other components that enable weapons to function, including the arming of autonomous vehicles. It strengthens language against tracking people without consent, using AI in criminal-justice decision-making, and building surveillance tools, while carving out exceptions for uses such as fraud monitoring and journalism. New standards were added for high-risk applications in health, finance and law, and for autonomous hardware, which the policy now says requires a qualified human operator and a safe-state fallback. A new clause prohibits sustained, needless cruelty toward Claude models in only the most extreme cases, which Anthropic said complements the model’s existing ability to end abusive conversations; the policy also clarifies enforcement against entities incorporated, headquartered or majority-owned in regions Anthropic does not support.

The update is a routine, scheduled revision rather than a response to a single incident, but it shows a usage policy still being rewritten in both directions at once — tightened on deception and weapons, loosened on political-speech restrictions its own users had found overbroad — as the range of real-world misuse and legitimate use cases the company has observed continues to grow.