New RedAgents AI Platform Deploys Multiple Hacking Agents to Find and Exploit Vulnerabilities

RedAgents, a cloud platform that deploys multiple AI penetration testing agents against authorized targets to discover vulnerabilities and test attack paths.

Announced on October 16, 2026, the service brings Artex Pro, Cyber(YanJi), Yanxun, and CyberStrikeAI Enterprise Edition into a single environment for security operations teams.

The platform addresses a limitation of automated security assessments: different agents can uncover different weaknesses, leaving organizations with incomplete visibility when they rely on a single testing engine.

Delivered as software as a service, RedAgents allows teams to run agents in parallel against the same target, select combinations for specific engagements, or deploy an individual agent.

Vulnerability findings, execution logs, and assessment data appear in a unified interface, reducing the need to consolidate results across platforms.

“No single agent can discover every issue,” ThreatBook founder and CEO Feng Xue said in the announcement. The company argues that combining agents improves vulnerability coverage while reducing false positives.

However, the launch announcement provides no independent comparative benchmark establishing the platform’s aggregate detection improvement or false positive rate.

Its technical proposition is therefore diversity: agents with different search strategies and operational strengths examine the same attack surface, potentially identifying paths that another agent misses.

ThreatBook says this model supports more frequent assessments as AI penetration testing capabilities evolve.

Artex Pro extends an autonomous penetration testing project developed by the RedAgents product lead.

The enterprise edition emphasizes execution control, operational stability, and traceable results alongside autonomous vulnerability discovery and exploitation.

ThreatBook observed RedAgents, as an orchestration layer that combines these approaches rather than requiring customers to manage separate tools and vendor integrations.

RedAgents AI Platform

Cyber(YanJi), a Python based agent, searches unknown state spaces for a viable route from an initial position to a defined success condition.

ThreatBook reports that it solved all 54 challenges in the Tencent Cloud Hackathon AI Penetration Testing Challenge, ranking first among individual participants.

Yanxun, developed by Changsha Yanzhi Technology from research associated with Peking University’s Changsha institute, targets extended penetration scenarios.

According to ThreatBook, it coordinates dozens of specialized agents and supports reconnaissance, threat intelligence collection, code auditing, penetration testing, and red team assessments. Its claimed reproducibility and zero false positives remain vendor assertions.

CyberStrikeAI Enterprise Edition adds native MCP support and integrations with more than 100 security tools.

ThreatBook acquired CyberStrikeAI in September 2026, connecting that acquisition directly to its broader autonomous testing strategy.

The enterprise version converts assessment requirements into controlled, traceable tasks within authorized scopes.

The launch also carries a significant dual use backdrop. ThreatBook says the original Artex project became closed source on October 8 after attackers misused it in financial sector incidents.

Separate reporting linked suspected abuse to attacks against South Korean financial institutions.

RedAgents addresses authorization through target allowlists that define testing boundaries.

ThreatBook says assessment data remains inside its environment rather than being transferred to third party agents, while Zero Data Retention support enables complete deletion of test data.

ThreatBook plans to add further open source agents alongside enterprise offerings. Its product page describes availability as limited testing, distinguishing the public launch from unrestricted general access.

The immediate proposition is consolidated offensive validation, backed by vendor stated controls rather than independently demonstrated security guarantees.

Stops Cyber threats before impact with 21 min faster MTTR. Integrate ANYRUN’s Sandbox in your SOC.