Agentic Browser Attack Surface Mapper

Produces a defensible threat model for browser-native AI agents (ChatGPT Atlas, Claude in Chrome, Perplexity Comet, OpenAI Operator, Opera Neon) before enterprise rollout — maps DOM-manipulation vectors, cross-origin data leakage, session/cookie theft paths, confused-deputy risks across tabs, phishing surface expansion, and indirect prompt-injection via rendered web content; outputs a prioritized mitigation plan, deployment guardrails, user-facing warnings, SOC detections, and an attestation checklist. Built for 2026, when every major AI lab shipped a browsing agent and prompt-injection remains an unpatched class of risk.