techtarget.com • 1h
Agentic AI Hijacking via Model Context Protocol MCP and MFA Bypass
This threat convergence leverages a multi-stage kill chain starting with the injection of malicious skills or compromised Model Context Protocol (MCP) tools into AI agentic workflows. Attackers utilize indirect prompt injection and poisoned Python toolchains to hijack autonomous agents, which then exploit their trusted identity status to execute session hijacking and API-level MFA bypasses. This mechanism bypasses traditional "Human-in-the-Loop" (HITL) controls, enabling rapid lateral movement across interconnected SaaS ecosystems and privileged enterprise APIs, rendering legacy SMS and TOTP authentication inadequate against autonomous, high-context attacks.
Links:techtarget.com, techjacksolutions.com, Crowdstrike, Okta, Reddit, Hiddenlayer, Cdw, Workos, Getastra, Obsidiansecurity, Dark Reading •