NORAD.io
OVERVIEWCOUNTRIESBOTSTHREATS

CLAUDE-WEB

LOW RISK🤖 AI ASSISTANT BROWSER

Anthropic's dedicated web browsing agent for Claude's tool-use web access

ORGANIZATION
Anthropic
FIRST SEEN
2024-06
RESPECTS ROBOTS.TXT
✓ YES
DOCUMENTATION
docs.anthropic.com
—
DAILY VISITS
—
COUNTRIES ACTIVE
TRACKING
STATUS
—
LAST SEEN

📡 CLAUDE-WEB USER-AGENT STRING

Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; Claude-Web/1.0; +https://anthropic.com/claude-web)

This is the User-Agent header sent by Claude-Web in HTTP requests. Use this to identify Claude-Web in your server access logs.

📋 ABOUT CLAUDE-WEB

Claude-Web is Anthropic's dedicated web browsing agent, used when Claude performs web access through its tool-use capabilities. This bot represents the evolution of Claude's web interaction from simple page fetches to more sophisticated browsing patterns that can follow links, extract specific information, and navigate multi-page research tasks.

Claude-Web is part of the growing ecosystem of AI agent browsers that go beyond simple URL fetching. In agentic workflows, Claude-Web may perform sequences of related page requests — searching, following links, and extracting information across multiple sources to complete research tasks requested by users.

NORAD.io monitors Claude-Web activity as part of its AI agent tracking capabilities. As AI assistants become more capable web agents, understanding their browsing patterns becomes critical for site operators. NORAD distinguishes Claude-Web (real-time browsing) from ClaudeBot (training crawling) to enable granular access control policies.

🎯 HOW TO DETECT CLAUDE-WEB

  • ▸User-Agent contains 'Claude-Web'
  • ▸Sporadic request pattern tied to user conversations
  • ▸May perform multiple related page fetches within a short time (following links in context)
  • ▸Distinct from ClaudeBot (training) and Anthropic-AI (earlier browsing UA)
  • ▸Increasingly common as Claude's tool-use capabilities expand

🔄 CRAWL BEHAVIOR

On-demand page fetching during Claude conversations and agentic workflows. Single-page requests driven by tool use. Does not systematically crawl. May follow links within a conversation context.

PURPOSE

Enables Claude to browse the web during conversations and agentic tasks. Used for real-time information retrieval, fact-checking, and research assistance. Data is not stored for training.

🤖 ROBOTS.TXT CONFIGURATION

User-agent: Claude-Web
Allow: /

# Claude-Web is the browsing agent, not the training crawler
# To block training: block ClaudeBot instead

Claude-Web respects robots.txt directives. Add this to your robots.txt file at the root of your domain.

→ Complete Guide: robots.txt for AI Bots

🗺️ WHERE IS CLAUDE-WEB ACTIVE?

🇺🇸
United States
🇨🇦
Canada
🇲🇽
Mexico
🇬🇹
Guatemala
🇨🇺
Cuba
🇭🇹
Haiti
🇩🇴
Dominican Republic
🇭🇳
Honduras
🇳🇮
Nicaragua
🇸🇻
El Salvador
🇨🇷
Costa Rica
🇵🇦
Panama

⚠️ RELATED THREATS

Prompt Injection

Attempts to override bot instructions via malicious content embedded in web pages

Data Exfiltration

Bots attempting to extract sensitive data from websites including PII and credentials

Credential Stuffing

Automated login attempts using leaked credentials from data breaches

Aggressive Content Scraping

Bots aggressively scraping content beyond robots.txt limits and ToS

🔗 RELATED BOTS

ClaudeBotLOW

Anthropic · Anthropic's web crawler for training Claude AI models

Anthropic-AILOW

Anthropic · Anthropic's AI assistant web browsing capability for real-time information retrieval

📂 MORE 🤖 AI ASSISTANT BROWSERS

ChatGPT-UserOpenAIPerplexityBotPerplexity AIAnthropic-AIAnthropicPerplexity-UserPerplexity AI

📚 RELATED GUIDES

How to Detect AI Bots →robots.txt for AI Bots →NORAD API Docs →

PROTECT YOUR WEBSITE

Deploy SiteTrust to monitor and control AI bot access to your site with the Agent Passport Standard.

INSTALL SITETRUST →