Empirical Security Analysis of Production AI Agent Frameworks, Cloud Data Pipelines, and Multimodal Trajectory Poisoning: OAuth Token Exfiltration, SSRF, and Pre-Tool Cryptographic Sandboxing

An empirical offensive security and vulnerability analysis of production AI agent frameworks, Model Context Protocol (MCP) bridges, cloud data pipelines, and multimodal trajectory poisoning vectors conducted by HM Shihabuddin Jashory (Shihab). Covers real-world OAuth token exfiltration chains, Server-Side Request Forgery (SSRF), sibling path traversal bypasses, and deterministic Pre-Tool cryptographic sandboxing defenses.