Recent news of multiple South Korean banks having their networks breached has once again deepened the industry's alarm over AI attacks. Even when attackers are not technically skilled, they can use AI agents to cause a major security impact.
When an attack takes only one sentence of natural language, how should enterprises defend?
The best defense is offense. In the human era and the AI era alike, the most effective way to validate the exposure surface is penetration testing and red team testing: before real attackers arrive, use the most advanced and most capable red teams to run simulated attacks, and find the problems ahead of the attackers.
And AI is changing the way enterprises conduct penetration testing.
In the past, an enterprise penetration test ran on a cycle measured in years. To get more complete results and to avoid being locked into the capability ceiling of a single vendor, enterprises also had to commission several human red teams at the same time. The front stretched out and the cost rose.
The arrival of AI lets penetration testing shift from a long-cycle major exam to a short-cycle quiz. At the same time, excellent AI penetration agents keep emerging and evolving, and large enterprises and individual developers alike may produce the best agent. In the AI era, the best way for enterprises to conduct penetration testing is to connect to a cluster of AI penetration agents that keeps evolving and can be called at any time, and so to continuously obtain the most advanced AI penetration testing technology and the most objective penetration results.
Therefore, ThreatBook has done something unprecedented:
We have integrated some of the strongest AI red team agents in the industry onto a single platform, so that enterprises can call on them freely, just as they use multi-engine antivirus.
This is RedAgents.Ai, ThreatBook's multi-AI-agent penetration platform.
RedAgents.Ai is a software-as-a-service (SaaS) platform. It supports independent connection and parallel operation of all kinds of AI red team agents from universities, security vendors and open-source communities. Like human red teams, this group of the best agents is hard to rank by level, and each has a different approach, so the vulnerabilities and weaknesses each finds also differ. Enterprises need not worry about being locked into the capability ceiling of a single vendor. They need not procure from several vendors or switch between several back ends. With one instruction, they can call multiple differentiated AI penetration agents to complete automated penetration testing and can easily cross-validate capabilities, while costs stay controllable and resources are saved.
On data compliance, RedAgents.Ai commits that all test data is kept in the ThreatBook environment and does not flow to any third-party agent. Defenders can define the testing scope through an allowlist, and the platform acts only within the authorized scope. In addition, the platform provides Zero Data Retention (ZDR) support, which can delete test data completely.
RedAgents.Ai has now connected several enterprise-grade proprietary agents and open-source agents, including the ARTEX Pro autonomous penetration testing agent (closed-source Enterprise Edition), the Cyber (Yanji) AI automated penetration testing agent, the Yanxun cyber attack-and-defense agent and CyberStrikeAI Enterprise Edition:
Artex (Open-Source Edition) is an autonomous penetration testing agent for enterprise attack surfaces, and a personal project of the RedAgents.Ai product lead at ThreatBook. ARTEX Pro (Enterprise Edition) focuses on strengthening security controls. Taking an engineering approach, it builds the autonomous agent into a complete red team system that can not only break through defenses and find vulnerabilities autonomously, but also remain controllable throughout, run stably and produce traceable results.
On September 3, 2026, Artex (Open-Source Edition) won the championship at Baidu Security's "Agent+" Attack-and-Defense Capability Challenge. Artex (Open-Source Edition) was previously operated as an open-source project. After it was maliciously exploited by attackers, triggering several large-scale financial industry security incidents abroad, Artex (Open-Source Edition) announced on October 8 that it was officially moving to closed source. Also, considering that Artex (Open-Source Edition) lacks built-in security controls, it is recommended for cautious use against test environments only.
Developed by the chainreactors team, it is the industry's first AI-native automated penetration system. Unlike the traditional "large model + tool orchestration" approach, Cyber (Yanji) models penetration testing as a directed search in an unknown state space. An "intent-fact" reasoning chain drives the test path, and multiple Workers explore in parallel on a shared blackboard and generate tasks dynamically, with no preset roles or fixed workflow. At the second Tencent Cloud Hackathon (TCH) Intelligent Penetration Challenge, Cyber (Yanji) achieved third place overall and first place individually, and was the only participant in the whole field to solve all 54 challenges.
An intelligent network penetration product from Changsha Yanzhi Technology Co., Ltd., built on the transfer of Peking University research results. It can autonomously execute the complete penetration testing process, from asset reconnaissance, vulnerability discovery and exploitation verification to post-exploitation. Yanxun uses AI agents as its execution core and can schedule dozens of specialized agents at the same time to work in parallel, compressing a detection cycle of several weeks into a few hours, and it ensures through a closed-loop vulnerability data verification mechanism that every vulnerability is reproducible with zero false positives. Yanxun supports multiple security assessment scenarios, including information gathering, threat intelligence collection, code auditing, penetration testing and red team assessment.
CyberStrikeAI is an open-source, AI-native autonomous penetration testing tool with native MCP support and integrations with 100+ security tools. It earned over 7,800 GitHub stars within just a few months of launch.
CyberStrikeAI Enterprise Edition turns security testing requirements into controlled, traceable automated tasks, efficiently orchestrating AI-powered penetration testing and red team exercises within clearly defined, authorized scopes. CyberStrikeAI was acquired by ThreatBook in September 2026.
RedAgents.Ai can fully meet enterprises' needs for routine penetration testing, attack-and-defense exercises and security compliance verification. In the future, RedAgents.Ai will keep connecting more differentiated, scenario-specific AI red team agents, enriching the roster of red teams on the platform. It will keep connecting university research results, new open-source works from the community and frontier agents from vendors, so that defenders can experience the latest AI attack-and-defense capabilities without any development or additional procurement.
ThreatBook hopes to use an open ecosystem to break down the capability silos in the field of AI penetration, and to provide red team capability that is truly independent and controllable, and flexibly orchestrated. What defenders truly need is the strongest red team squad, one that can be "called up by name" at any time and combined on demand.
RedAgents.ai is now officially live and open for limited testing.
Request a trial now and use the industry's strongest AI agent penetration cluster to assess your own exposure surface before attackers do.