The Claude in Chrome browser extension uses a server-side security list to restrict user interactions on specific websites. When a user attempts an action, the extension sends the page URL to the Anthropic server, which returns a category code determining the level of permission required.
The server assigns categories that dictate behavior: category0 allows permanent permissions, category1 blocks access entirely, and category3 requires manual confirmation for every individual click. These categories are applied at the domain level rather than the page level, and the server response is cached for 5 minutes.
Users cannot manually override these restrictions or add sites to an approved list within the personal version of the extension. The option to manage trusted sites is restricted to Team and Enterprise plans. Anthropic maintains these security measures to mitigate risks associated with prompt injection attacks.
Developers can inspect the category assigned to a specific site by using the Chrome Developer Tools. By monitoring the network request to the url_hash_check endpoint, users can view the assigned category in the server response. Anthropic does not publicly disclose the criteria for site categorization.
