Anthropic's updated usage policy, announced on October 8 and effective November 12, adds four substantive prohibitions, loosens one election rule, and turns two long-standing recommendations for high-risk deployments into requirements. The headline about users being "mean to Claude" is the least consequential part for anyone running the model in production.
What is new
Sustained abuse of the model. The policy now prohibits "sustained and needless abusive or cruel behavior toward our models." The scope is deliberately narrow: it applies "only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose," and explicitly excludes "common versions of user frustration, pushback, dark creative themes, or model testing and research." Enforcement is the existing conversation-ending feature; Anthropic says this "will remain the primary enforcement mechanism." No warning or suspension ladder is described.
Deceptive campaigns. A new section, "Do Not Engage in Deceptive Campaigns or Artificial Activity," covers efforts to "obscure who is behind a message" and building "the tools and infrastructure for running influence operation campaigns." TechCrunch's examples are fake accounts and fabricated news outlets.
Weapons and drones. The prohibition now explicitly covers "software and components that make weapons work" and "arming drones and other autonomous vehicles." The Register reports this followed attempts to build guidance and control systems for weapons, including armed drones.
Surveillance. Tracking people without consent is prohibited. Claude "cannot be used to decide or recommend who to investigate, arrest, or charge," nor to build or improve surveillance tools.
What is loosened
The elections section was renamed "Do Not Undermine Democratic Processes" and narrowed to deceiving voters and disrupting elections. The former blanket ban on personalised vote and campaign targeting was removed. Deceptive targeting and misuse of personal data remain prohibited. This is the detail most coverage skipped.
What now binds deployments
For high-risk uses, which Anthropic lists as health, legal rights, finances, livelihood and access to essential services, the policy requires "a qualified 'human in the loop'," defined as someone "who has the authority to review and, if necessary, change Claude's recommendations," and requires that the affected individual be told that AI was used. For Claude connected to physical equipment, operators must be able to "observe the equipment and stop it if needed," and "the equipment must also be able to hold a safe state if Claude is disconnected."
The policy does not distinguish API customers from consumer users; it is written as a single document. Anthropic frames most of the changes as clarifications of existing rules.
The one critique worth noting
The Register's Carly Page points out that "it's not clear how Anthropic intends to distinguish between legitimate frustration and gratuitous cruelty," and that there is no established evidence the model experiences distress. Fair, but since the rule rests on a feature that already exists and the exclusions are broad, the practical risk to business users is close to zero.
What to do before November 12
If your product uses Claude to screen loan applications, rank job candidates, triage insurance claims or anything else on the high-risk list, two things now have to be true and documented. A named person with authority to override the model's recommendation must sit in the process, and the person affected must be told AI was involved. In our own client work the override step usually exists but is undocumented, and the disclosure usually does not exist at all. Both are small changes: write down who approves what, and add one sentence to the screen the end user sees. The same two steps are what most data-protection regimes ask of automated decision-making, so the work is not wasted on a single vendor's policy.
Sources: Anthropic, 2026 usage policy update, TechCrunch, The Register
