Palo Alto Networks’ Unit 42 is significantly expanding its Frontier AI Defense capabilities through a new partnership with Armadin, an AI-native offensive security company founded by former Mandiant CEO Kevin Mandia. The partnership integrates Armadin’s autonomous attack platform directly into Unit 42’s elite service, allowing for continuous, AI-driven validation of external attack surfaces against sophisticated frontier AI threats.
As frontier AI models become more capable, the threat landscape is shifting rapidly. Attackers now have access to tools that enable faster reconnaissance, more creative exploit chaining, and fully autonomous campaign execution. Traditional red teaming, which depends heavily on human experts conducting periodic exercises, is struggling to keep pace with adversaries who can deploy their own AI agents around the clock. Unit 42 originally built Frontier AI Defense to counter this evolution by using advanced AI models to proactively map likely attack paths and help organizations remediate high-risk exposures before they are exploited in the wild.
The addition of Armadin takes this defense to the next level. Instead of relying solely on human-directed analysis, the service now incorporates a fully autonomous offensive layer capable of simulating real attacker behavior at machine speed. This partnership reflects the broader industry move toward agent-versus-agent security testing, where defensive systems are no longer evaluated only through static reports but through continuous, dynamic simulation that closely mirrors how modern threats actually operate. For security teams responsible for protecting complex environments, this represents a meaningful step toward more realistic and timely validation of their posture against frontier AI-powered attacks.
Key Terms
Frontier AI Defense: Unit 42’s specialized service that uses advanced AI models to proactively identify and help remediate high-risk exposures from frontier AI-powered threats.
Autonomous Attack Validation: The use of AI agents to continuously and independently simulate sophisticated attacks against an organization’s external environment.
Agentic Offensive Security: Security practices that leverage autonomous AI systems to emulate advanced attacker tactics, techniques, and procedures at scale.
External Attack Surface: All internet-facing assets, applications, and entry points that adversaries can target from outside the organization.
Armadin: AI-native offensive security company founded by Kevin Mandia, focused on autonomous red teaming and attack simulation capabilities.
Conditions Driving This Partnership
The rapid advancement of frontier AI models is giving attackers powerful new tools for reconnaissance, exploit development, and fully autonomous campaign execution at speeds previously impossible for human operators.
Traditional periodic red teaming and manual vulnerability scanning can no longer match the persistence, creativity, and velocity of AI-driven attacks that operate 24/7 without fatigue.
Boards and executive leadership are demanding concrete, evidence-based proof that security controls can withstand real-world frontier AI threats, moving beyond theoretical risk assessments and compliance checklists.
Defensive AI systems themselves are becoming more sophisticated and autonomous, creating an urgent need for equally advanced offensive simulation tools to properly test and validate them.
Organizations face increasing difficulty in manually validating complex and constantly changing external attack surfaces, especially as cloud deployments, API integrations, and AI-powered applications expand the perimeter faster than security teams can keep up.
Regulatory and stakeholder pressure is growing for demonstrable resilience against AI-augmented threats, pushing companies to adopt continuous validation methods rather than point-in-time assessments.
The rise of agentic AI on both sides of the battlefield is forcing a fundamental evolution in offensive security — from human-led exercises to autonomous, agent-vs-agent simulation that better reflects the current and future threat environment.
There is a clear market gap for integrated solutions that combine elite human expertise with scalable AI-driven attack simulation, allowing security teams to test defenses at the same speed and autonomy that adversaries now possess.
What Frontier AI Defense Looked Like Before
Before partnering with Armadin, Unit 42’s Frontier AI Defense relied on a powerful combination of highly skilled consultants and frontier AI models to analyze potential attack paths and deliver strategic remediation guidance. These assessments were among the most advanced available, providing organizations with detailed reports on likely vulnerabilities, attack chains, and recommended fixes. However, the process was still largely human-directed and conducted on a scheduled, periodic basis. While extremely valuable, this approach created inherent limitations.
Between assessments, new exposures could emerge as environments changed due to cloud migrations, new application deployments, or configuration updates. Attackers leveraging their own AI tools could move faster and more creatively than defenders could validate their posture in real time. Many security teams were left with reasonable confidence based on expert analysis and modeling, but they often lacked frequent, dynamic, real-world proof of how their defenses would actually perform against persistent, autonomous, and highly capable AI-driven campaigns. The gap between theoretical risk modeling and live exploit validation was a persistent challenge in keeping pace with the evolving threat landscape.
What’s Changing Now
The integration of Armadin’s autonomous attack platform fundamentally changes how Unit 42 delivers Frontier AI Defense. Unit 42 can now run continuous, AI-powered offensive simulations as a core component of the service. Armadin’s agents autonomously probe external perimeters, discover and chain exploits, and validate real-world exploitability at machine speed and scale. This creates a much tighter and more realistic feedback loop: potential exposures are identified, actual attack scenarios are simulated in real time, successful compromise paths are proven with evidence, and prioritized, actionable remediation recommendations are delivered based on what truly works against the current environment.
Customers benefit from significantly more frequent and dynamic testing without a proportional increase in human effort or cost. The combined solution is already commercially available and gives organizations one of the most advanced capabilities on the market today for continuously testing their resilience against frontier AI threats. This partnership positions Palo Alto Networks and Unit 42 as leaders in bridging the critical gap between defensive AI systems and realistic, autonomous offensive validation — moving the industry closer to true agent-versus-agent security testing.
Our Take
AI Security Take
This partnership between Unit 42 and Armadin represents a meaningful evolution in how organizations can defend against frontier AI threats. By combining world-class human expertise with fully autonomous AI-driven attack simulation, they are shifting from static, periodic assessments toward continuous, realistic validation of external attack surfaces. In an era where both attackers and defenders are rapidly becoming more agentic, the ability to test defenses with comparable autonomy, speed, and persistence is quickly becoming essential rather than optional.
For governance, risk, and security teams, this development highlights a critical shift happening right now: periodic audits, policy documents, and theoretical modeling alone are no longer sufficient to demonstrate credible security posture. Real-time evidence generated from dynamic attack simulation is emerging as a core requirement. If your organization is concerned about frontier AI risks or wants to validate its defenses against autonomous, AI-powered threats, head to the GAIG marketplace right now. Compare the platforms that deliver advanced runtime monitoring, adversarial defense testing, agentic security capabilities, and continuous validation so you can move from theoretical risk management to proven, evidence-based resilience.
