Story
October 9, 2026
Anthropic’s Claude crackdown puts AI safety—and AI personhood—on a collision course
Anthropic’s revised Claude rules tighten curbs on election deception, weapons and surveillance while allowing legitimate civic uses. But its narrow ban on cruelty toward the chatbot has reignited a fight over whether model welfare goes too far.
Anthropic’s updated Usage Policy, due to take effect on November 12, arrives as Claude is being asked to perform longer and more independent tasks. The company says the rewrite mostly clarifies existing enforcement, but it also gathers scattered restrictions into sharper lines around political manipulation, weapons and surveillance.1
First, Anthropic consolidated rules against deceptive political and commercial campaigns, including fake-account networks, concealed sponsors and fabricated content. It also narrowed its election language: Claude may still support legitimate civic work, such as translating voter information or helping officials send ballot-cure notices, but cannot be used to mislead voters, impersonate officials or suppress turnout.1
The policy then makes explicit that its weapons ban covers not only weapon design but “software and components that make weapons work,” including arming drones and other autonomous vehicles. On surveillance, it bars non-consensual tracking and using Claude to decide whom law enforcement should investigate, arrest or charge, while preserving uses such as journalism, legal research, content moderation and consent-based fraud monitoring.1
The flashpoint is a new prohibition on “sustained and needless abusive or cruel behavior” toward Claude. Anthropic insists it applies only to repeated, purposeless extreme conduct—not ordinary frustration, dark creative work or model testing—and says Claude ending rare abusive conversations remains the main enforcement tool.1 The Verge reported that Anthropic had not said whether penalties could extend beyond ending a conversation.2
That caveat has not quelled the backlash. In a reposted critique, David Sacks amplified the argument that Anthropic was “anthropomorphizing” its models by linking the ban to uncertainty over Claude’s moral status.
3
Yet Anthropic is simultaneously pitching Claude as a defensive tool, not merely an entity to be protected. Its Critical Infrastructure Defense Program offers frontier models, engineers and threat research to firms protecting power grids and water systems, although questions remain about safely deploying AI-suggested fixes in live utility environments.4