Picus Security has introduced a new capability titled Picus Autonomous Penetration Testing, designed to integrate artificial intelligence agents into offensive security operations. This launch marks a transition for the company from traditional automated testing toward an agentic model capable of simulating adversary behavior in production environments. The platform is built upon five years of data gathered from enterprise deployments, aiming to address the increasing velocity of vulnerability disclosures, which saw 36,971 CVEs published in the third quarter of 2026 alone.
The core functionality of the system involves AI agents that perform reconnaissance, vulnerability assessment, and initial access, followed by post-exploitation maneuvers. Unlike static testing tools, these agents are designed to chain multiple exploits together to identify potential attack paths toward an organization's most critical assets. The system is intended to operate continuously, providing a countermeasure to the rapid pace at which modern adversaries identify and weaponize new vulnerabilities.
Picus Security has implemented several governance controls to ensure the autonomous agents do not compromise production stability. Before any testing begins, organizations must define the scope and permitted techniques. The platform includes hard-coded restrictions that prevent the agents from performing destructive actions, regardless of the AI's decision-making process. Additionally, users can select from three operational intensity levels—Cautious, Standard, or Elevated—and retain the ability to terminate all agent activity with a single command.
The system offers two distinct modes of execution. In fully agentic mode, the AI explores the environment dynamically to discover novel attack vectors. Alternatively, users can opt for a deterministic mode that relies on a pre-tested library of techniques, providing a more predictable outcome for sensitive production environments. The platform does not require new software deployments, as it utilizes the existing Picus agents already present within customer networks and operates independently of Active Directory.
The output generated by these tests consists of documented attack paths that demonstrate how an adversary could achieve specific impacts, such as credential theft, domain takeover, or unauthorized access to sensitive data. These findings are integrated into a centralized, deduplicated backlog alongside data from the company's existing Breach and Attack Simulation and Exposure Validation products. This consolidation is intended to help security teams prioritize remediation efforts by showing not only what an attacker can exploit but also which security controls successfully detected or blocked the individual steps of an attack chain.
Picus Security emphasizes that the engine is designed to balance aggressive exploration with the operational requirements of enterprise software. By utilizing historical data from hundreds of previous customer deployments, the company claims the system can mimic the creativity of a human red team while maintaining the discipline required for continuous, automated operation. The platform is positioned as a response to the limitations of point-in-time penetration testing, which the company argues is increasingly insufficient in an environment where adversaries operate at machine speed.





