Claude Code and other local AI agents are now named, not flagged
Agents that run on a person's own computer, such as Claude Code, now appear under their own name with an honest 'cannot be verified' status instead of being counted as failed identity checks.
Some AI agents no longer fetch pages from their maker's servers. Claude Code, Anthropic's coding agent, runs on a developer's own machine and announces itself as Claude-User (claude-code/…) when it reads a page. Because the request comes from the developer's network, it can never match Anthropic's published IP ranges, and until now Salience recorded it as a bot that claimed to be Claude-User and failed every identity check. On most sites that was the majority of Claude-User traffic.
Those requests now appear as Claude Code, with a verification status of cannot be verified rather than unverified. The difference matters: unverified means a check exists and the request failed it, which is what impersonation looks like; cannot be verified means no check exists for this agent. Claude Code no longer counts towards the "Unverified Bots Claiming Identity" alert or the impersonator share in the bot access table, and it has its own entry in the crawler filters.
The same treatment applies to other user-triggered and index agents we were lumping under generic names: Manus-User, MistralAI-User, OpenAI MCP, ExaUser, ExaSearchBot, Meta-ExternalFetcher and YouBot.
The AI page fetches report now shows these declared agents by default, with their own Declared evidence label and a Declared agents total, instead of hiding them behind the Include unverified requests switch. That switch now means exactly what it says: requests that claimed a verifiable agent and failed the check.
One thing to know: agents such as Codex, Cursor and most browser-driving agents send no identifying name at all. Their fetches arrive looking like curl, python-requests or a plain HTTP client, so they still appear under those names. See AI page fetches in the help.