Vendor Combines AI Attack Agents, Human Experts to Simulate Real-World Cyberattacks
A new player has entered the cybersecurity landscape, making headlines with a significant funding achievement and innovative approach to penetration testing. The startup, Armadin, has recently emerged from stealth mode, securing nearly $190 million to revolutionize red-teaming practices through the use of autonomous artificial intelligence agents.
Led by co-founder and Chief Offensive Security Officer Evan Pena, Armadin seeks to transform how organizations assess their security postures. This funding, which comes from both Seed and Series A rounds spearheaded by Accel, will enable the San Francisco-based company to develop AI agents that are meticulously trained in expert red-team methodologies. These agents aim to continuously test environments at scale, effectively mimicking complex offensive security workflows.
Pena remarked on the transition from traditional cybersecurity tools, stating, “A lot of what you were seeing was mostly around chatbots and large language models taking in a lot of text and being able to give you advice and reporting.” However, he highlighted a paradigm shift where AI can now not only engage in conversation but also orchestrate tasks and execute them. This growing sophistication ultimately enhances the depth and effectiveness of security testing.
The leadership of Armadin includes Kevin Mandia, who boasts a rich history in threat intelligence and incident response. Mandia previously spent two decades at Mandiant, a firm he helped elevate to prominence before selling it to FireEye in 2013. After serving as FireEye’s CEO and overseeing several significant acquisitions, Mandia left Google in 2024 after Mandiant was purchased for a substantial sum.
One of the most groundbreaking elements of Armadin’s strategy lies in agent-to-agent communication protocols. This innovative aspect allows multiple AI entities to coordinate simultaneous attacks across various domains, such as web applications, external infrastructure, and internal networks. In contrast to traditional penetration testing—which often focuses on a single attack path—AI-powered systems are capable of uncovering dozens or even hundreds of potential vulnerabilities within a short span of time. This capability fundamentally changes the dynamics of security assessments.
Pena noted, "What we’re doing right now is two things on the red-team side. The first one is going to be training the agents to do what it is that we do, the way we do it." This involves human red-teamers encoding their expertise into formats that AI can leverage. By documenting methodologies, tools, and decision-making processes, these experts provide valuable context that the AI systems can utilize.
Notably, Pena acknowledged the challenges of using generic AI models for offensive security. "A frontier model out of the box, if you tell it, ‘Hey, go hack this thing,’ it may just not work," he explained. It is essential for the AI to undergo fine-tuning and have access to a specialized toolkit that enables it to execute tasks effectively. Human red-teamers play a critical role in adding nuance and precision to the AI’s capabilities.
This automated approach to security testing introduces a transformative shift in vulnerability assessments. Human teams often take the path of least resistance, concluding engagements after identifying one viable attack route. In contrast, Armadin’s AI-driven systems can continually probe and explore a multitude of attack paths, providing a more comprehensive security evaluation.
Pena emphasized the stark difference that this AI approach offers, citing, "I’ve been doing this my whole career, assessing security networks across one enterprise. We’re almost always successful. And the reason why is just because it’s always been human-led." The ability to conduct ongoing assessments rather than an annual review could lead to enhanced security coverage and a deeper understanding of potential vulnerabilities.
Looking to the future, Armadin plans to use AI agents to run continuous attacks that identify vulnerabilities while allowing human experts to validate complex findings. This collaborative approach aims to integrate human intuition with AI efficiency, enriching the security landscape. As human teams discover new strategies, these can be seamlessly added to the AI’s playbook, continually enhancing the system’s effectiveness.
Addressing the challenges of vulnerability remediation is another critical focus for Armadin. The company hopes to reduce the bottlenecks that often arise in this process by identifying vulnerabilities that pose the greatest business risk and prioritizing remediation based on potential impacts. While AI may eventually facilitate automated remediation, Pena is cautious about its implications for operational stability, outlining the need for human oversight in critical areas.
In conclusion, Armadin aspires to enhance organization security through tactical and strategic recommendations aimed at bolstering defenses against evolving threats. By orchestrating multiple attacks throughout the entire attack lifecycle, Armadin’s AI agents can rapidly assess whether a firm’s detection systems are effectively identifying malicious behavior. This approach holds the promise of generating actionable insights—thus enabling organizations to strengthen their security architectures. As Pena noted, the integration of human and AI capabilities is not only about efficacy but also about evolving the cybersecurity landscape in a meaningful way. As Armadin continues its journey, it positions itself at the convergence of innovation and security, heralding a significant evolution in how organizations protect themselves in an increasingly complex digital world.