Blog

Blast Radius Tiers: A Permission Model for Agents That Act

Anthropic found users approve 97% of permission prompts while human review catches 13.6% of dangerous commands. Approval was already theatre. Here is the tier model that replaces it.