What happened
On Oct 21, 2026 Escape launched AI Red Teaming inside its AI Pentesting platform: Cascade now attacks MCP servers through the chatbot that fronts them (or pentests the MCP server directly over its JSON-RPC protocol), performs multi-turn jailbreaking, and produces exploit-level evidence rather than odd model answers.
Why it matters
It tests the crossover between the frontend, agent, and MCP server that point tools miss, and validates prompt-injection impact with real exploits — a meaningful step for agent and MCP runtime security.
Applicability
AppSec/pen-test teams shipping AI agents or MCP servers should evaluate it for agent-to-MCP traffic and multi-turn jailbreak coverage; useful to pair with static guardrail checks.