Summary
- Picus Autonomous Penetration Testing combines AI-directed exploration with established techniques to identify and validate attack paths.
- Organisations define testing scope, permitted techniques and operational limits, with restrictions intended to prevent destructive activity.
- The capability is available through the company's existing platform, but its fully agentic mode remains in early access.
Enterprise security teams can now use AI-directed penetration testing to investigate how attackers might combine weaknesses across their infrastructure, following an expansion of Picus Security’s automated assessment platform.
Picus Security introduced Autonomous Penetration Testing on 12 November, adding adaptive testing capabilities to its existing exposure validation offering. The system is designed to investigate possible routes through an enterprise environment rather than assess vulnerabilities solely as separate technical findings.
The capability combines AI agents with a library of established testing techniques, allowing organisations to examine whether weaknesses can be exploited and connected into attack paths towards sensitive systems.
Picus says the core functionality is available to customers through its Autonomous Exposure Validation Platform. Its fully agentic mode, in which agents independently select successive testing actions, remains available through an early access programme.
Testing follows the path of an intrusion
Traditional vulnerability assessments identify weaknesses in software, configuration and exposed infrastructure. Penetration testing goes further by examining whether those conditions permit an attacker to gain access or reach a particular objective.
Picus’s new capability attempts to combine those activities through an adaptive assessment process. Its agents examine the available environment, select potential techniques and reconsider their approach as each test produces additional information.
The company describes a sequence that can include reconnaissance, vulnerability assessment, initial access and post-exploitation within authorised boundaries.
A vulnerability that appears limited in isolation may become more significant when combined with another weakness. For example, access to a poorly protected system could expose credentials or network connections that provide a route towards infrastructure containing sensitive information.
The assessment is intended to establish whether such a route is technically possible, rather than merely assign a severity score to each underlying weakness.
Picus says successful testing steps are recorded as evidence supporting the resulting attack path. The findings are intended to distinguish demonstrated exposure from vulnerabilities whose practical exploitability remains uncertain.
That distinction affects remediation priorities. A weakness providing an attacker with access to sensitive infrastructure may require more immediate attention than another vulnerability with a higher theoretical severity but no demonstrated route to the affected service.
However, a successful test establishes what was possible under the conditions of the assessment. It does not demonstrate that an actual attacker has followed the same path or that every potential intrusion route has been identified.
Operating limits restrict testing behaviour
Introducing autonomous decision-making into penetration testing creates responsibilities beyond those associated with passive scanning.
Active assessments can interact with production applications, authentication systems and network services. Even where the activity is authorised, unexpected behaviour may affect availability or interfere with legitimate operations.
Picus says customers must approve the scope and permitted techniques before testing begins. Restrictions are enforced through rules that the AI agent cannot override, including prohibitions on specified destructive actions.
Administrators can select Cautious, Standard or Elevated operating settings, establishing different boundaries around the techniques available during an assessment.
The platform also provides a stop control and records testing activity for audit purposes. These controls are intended to allow organisations to monitor the assessment and interrupt it when necessary.
Two execution approaches are available. The deterministic mode uses techniques that Picus has previously tested, while the fully agentic mode allows more independent selection of successive actions.
The company describes its restrictions as suitable for production testing, although the announcement does not provide independent measurements establishing that the controls prevent every possible operational impact.
Organisations deploying such systems must therefore determine whether the permitted techniques are appropriate for the services being assessed, including applications that cannot tolerate unexpected interruptions.
Findings feed existing validation workflows
The new testing capability operates alongside Picus Breach and Attack Simulation and Exposure Validation.
Those functions assess related but different security questions. Penetration testing establishes whether an attack path can be completed, while breach and attack simulation examines how existing controls detect or prevent representative techniques.
Exposure validation provides further information about whether identified weaknesses can be exploited under the conditions present in an organisation’s environment.
Picus says the results are consolidated into a shared findings model and prioritised backlog, reducing duplication between separate assessment processes.
Connecting these results can help establish whether an exploitable weakness was detected, which defensive control failed or succeeded, and which corrective change would interrupt the demonstrated attack path.
That evidence still requires interpretation. A defensive control may respond differently to a technique that varies from the one tested, while a successful assessment does not establish the complete range of an attacker’s possible behaviour.
The accuracy of the resulting priorities depends partly on the coverage of the environment and whether testing reflects the configurations, permissions and dependencies present in production.
Picus says existing customers can use the capability through their deployed Picus agent, without adding a separate endpoint agent or relying on Active Directory integration.
That reduces some deployment requirements for current customers, although the scope of the assessment remains limited to the systems and assets accessible through the platform.
The addition of adaptive penetration testing extends the role of automated validation within enterprise security operations. Its ability to replace particular manual testing activities will depend on the reliability of its evidence, the range of attack paths it can assess and the controls governing its operation.
The core product is available to Picus customers, with the more autonomous execution mode continuing through early access.




