Three Ways AI Agents Operate Inside Your Browser
AI agents operate inside enterprise browsers through three distinct methods. Chrome DevTools Protocol (CDP), originally designed for developers, allows external applications to connect to debuggable browsers and control their tabs, giving agents access to authenticated sessions. Agents also embed directly in browsers as extensions like Claude in Chrome, or as standalone agentic browsers such as Perplexity Comet and ChatGPT Desktop. The third method, WebMCP, is an emerging web standard developed through the W3C Web Machine Learning Community Group with contributions from Google and Microsoft. Instead of simulating user interactions, agents can call tools that websites expose directly. Chrome introduced an early preview in 2026, and the specification continues to evolve.
A critical security gap exists: traditional endpoint, network, and application controls were built to see processes, traffic, and logins, not what an agent reads and decides inside an authenticated session. When agents act, applications record these as authenticated user requests and may not distinguish them from employee actions in audit logs. Agents that interpret web content face prompt injection risks from instructions embedded in websites, documents, or emails that attempt to redirect their behavior. WebMCP introduces additional trust problems: websites define what tools agents can invoke, and compromised or malicious sites could use misleading descriptions to manipulate agent behavior. Agents working across multiple sites can also move data between them, expanding potential data exposure.
Read the full article
Continue reading at BleepingComputer →This is an AI-generated summary. Read the original for the full story.