Artificial intelligence is changing the cybersecurity landscape, and its potential impact now extends beyond enterprise networks, applications, and cloud infrastructure into the physical world.
New research from Booz Allen highlights how advanced AI agents can navigate unfamiliar operational technology environments, adapt when their initial approaches fail, and interact with industrial equipment with limited prior knowledge of the environment.
In controlled laboratory testing, AI models completed eight scenarios, including one in which they progressed from a perimeter compromise to activity inside an industrial control network in just over 16 minutes. In another scenario, an AI model identified a robotic arm, analyzed its movement constraints, and moved it within minutes.
These findings highlight a critical concern for organizations responsible for manufacturing, utilities, transportation, and other essential services: AI can accelerate the path from a digital security compromise to a physical operational consequence.
What Did Booz Allen Discover?
Booz Allen evaluated advanced AI models in an isolated operational technology cybersecurity lab designed to represent a multi-vendor industrial environment.
The models were not provided with the environment’s source code, engineering documentation, or specialized OT guidance in advance. Despite this, they were able to investigate unfamiliar systems, identify equipment, research device information, develop attack plans, troubleshoot obstacles, and pursue the defined objectives.
The research demonstrated several important capabilities:
- Identifying unfamiliar industrial assets and gathering information about their configurations.
- Connecting technical findings to potential weaknesses in industrial systems.
- Adapting approaches when initial attempts were unsuccessful.
- Navigating pathways between enterprise-facing infrastructure and industrial control environments.
- Interacting with control systems and physical equipment when authorized in the test environment.
The study does not establish that every industrial facility is equally vulnerable or that an AI agent could automatically reproduce these results in a real-world environment. Actual outcomes depend on network architecture, equipment configurations, safety mechanisms, access privileges, and other defensive controls.
However, the research demonstrates why organizations should evaluate AI-assisted threats as part of their industrial cybersecurity planning.
Why Operational Technology Security Is Different
Operational technology refers to the hardware and software used to monitor, control, and manage physical processes.
Examples include:
- Industrial control systems (ICS)
- Supervisory Control and Data Acquisition (SCADA) systems
- Programmable Logic Controllers (PLCs)
- Industrial robots and automated machinery
- Manufacturing execution systems
- Building management and industrial monitoring systems
- Water, energy, and transportation control infrastructure
Unlike many conventional IT systems, OT environments directly influence physical operations.
A compromised business application may expose information or interrupt a service. A compromised industrial control environment could also affect production lines, equipment movement, environmental conditions, or the availability of essential services.
This makes the consequences of a cyber incident potentially much broader.
OT security must therefore consider not only confidentiality and data protection, but also safety, availability, equipment integrity, and operational continuity.
How AI Changes the Industrial Threat Landscape
Historically, navigating an unfamiliar industrial environment could require substantial technical knowledge, manual investigation, specialized engineering expertise, and time.
AI agents can potentially combine several of these tasks into a more coordinated workflow.
The Booz Allen research illustrates three developments that deserve particular attention.
1. Faster Discovery of Unfamiliar Systems
AI agents can help analyze available information about devices, firmware, network relationships, and control interfaces.
This creates a challenge for organizations that depend on the assumption that unfamiliar or poorly documented systems will be difficult for an attacker to understand.
Security through obscurity is not a reliable defensive strategy. Organizations need effective access controls, visibility, segmentation, and monitoring even when their systems are specialized or not widely known.
2. More Adaptive Attack Behavior
The research showed that the models could adjust their approach when an initial attempt did not work as expected.
This matters because defensive controls may block one pathway without eliminating other ways an attacker could reach a target.
Security teams should evaluate whether their controls work together across the full environment rather than assuming that a single firewall, endpoint tool, or access restriction will prevent every possible attack path.
3. A Shorter Window for Detection and Response
The reported time to reach an industrial control network in one test was just over 16 minutes following an initial perimeter compromise.
This does not mean every attack will proceed at that speed. It does demonstrate that defenders may have limited time to identify suspicious activity and intervene once an attacker has obtained an initial foothold.
Organizations should prioritize continuous monitoring, rapid escalation, preapproved containment procedures, and clear coordination between IT, OT, engineering, and safety teams.
When Cybersecurity Becomes a Physical Safety Issue
One of the most significant aspects of the research was the demonstration involving a robotic arm.
The AI model identified the equipment, mapped its protection zones and movement limits, and moved it within minutes in the controlled lab.
This highlights the relationship between cyber risk and physical safety.
Unauthorized changes to industrial control systems could potentially affect machine operations, production quality, process stability, or equipment availability. In certain environments, such changes could also create hazards for personnel.
The severity of any real-world consequence would depend on the system, the action attempted, the physical process, and the effectiveness of independent safety mechanisms.
Organizations should therefore ensure that cybersecurity controls complement, rather than replace, established industrial safety engineering.
Independent safety systems, local control capabilities, equipment interlocks, and emergency shutdown mechanisms remain important layers of protection.
Why Traditional Defenses May Not Be Enough on Their Own
The Booz Allen tests found that existing protections sometimes blocked or slowed the models. However, no single safeguard consistently prevented the models from reaching their objectives across all scenarios.
This does not mean that firewalls, network segmentation, endpoint protection, or access controls are ineffective. It reinforces the need for defense in depth.
A strong OT security program should combine multiple controls so that the failure of one layer does not automatically expose critical equipment.
Key priorities include:
Network segmentation: Separate enterprise IT networks from OT environments and restrict communications between zones according to operational requirements.
Secure remote access: Protect remote maintenance connections, engineering workstations, jump servers, and vendor access paths with strong authentication and carefully controlled permissions.
Least privilege: Limit users, service accounts, and connected systems to the access required for their specific operational roles.
Asset visibility: Maintain an accurate inventory of industrial devices, controllers, firmware versions, network connections, and system owners.
Configuration hardening: Disable unnecessary services, review trusted connections, eliminate inappropriate preauthenticated sessions, and secure administrative interfaces.
Operational monitoring: Detect unusual controller changes, unexpected engineering activity, configuration modifications, and abnormal communication patterns.
Independent safety mechanisms: Verify that safety interlocks and local controls remain effective even if supervisory systems or network-connected components are compromised.
What Organizations Should Do Now
The findings offer practical lessons for businesses that operate or depend on industrial technology.
1. Assess the Full IT and OT Attack Surface
Identify connections between corporate networks, cloud services, remote access systems, engineering workstations, and industrial control networks.
Pay particular attention to systems that bridge security zones, because they can create pathways between environments with very different risk profiles.
2. Test Realistic Attack Paths
Traditional vulnerability scans are useful, but they may not reveal how multiple weaknesses combine to expose an industrial process.
Authorized security assessments should examine network pathways, access permissions, exposed interfaces, remote administration, and the potential consequences of compromise.
Testing must be carefully planned for OT environments to avoid disrupting production or creating safety hazards.
3. Monitor Changes That Could Affect Physical Operations
Security teams should establish visibility into changes such as unexpected controller program writes, altered operating modes, unusual SCADA configuration imports, unauthorized engineering access, and unexplained equipment behavior.
Monitoring should be coordinated with operational teams so that suspicious activity can be distinguished from legitimate maintenance and production changes.
4. Prepare for Rapid Incident Response
Create incident response procedures specifically for OT incidents.
Define how security teams, plant operators, engineering teams, safety personnel, and leadership will communicate and make decisions when suspicious activity threatens operational processes.
Response plans should prioritize personnel safety and safe operating conditions. Abruptly disconnecting or shutting down industrial equipment can itself create risks if not handled according to operational procedures.
5. Validate Recovery and Restoration Capabilities
Maintain tested backups of industrial configurations, controller programs, SCADA projects, and other essential operational data.
Organizations should know how to restore trusted configurations and verify system integrity after a security incident.
Recovery exercises should include the technical and operational steps required to return systems to a safe, stable state.
6. Evaluate AI-Specific Threat Scenarios
Security teams should consider how AI agents might accelerate reconnaissance, adapt to defensive controls, and coordinate actions across connected systems.
Where appropriate, organizations should conduct controlled exercises to evaluate detection capabilities, response times, access restrictions, and the resilience of safety controls.
7. Establish Governance for AI and Automation
Organizations deploying AI agents internally should define which systems the agents may access, which actions require approval, and how their activities are logged and monitored.
High-impact actions involving industrial controls should be subject to strict authorization, safety validation, and human oversight appropriate to the potential consequences.
Industries That Need to Prioritize OT Security
The research has particular relevance to organizations where digital systems interact with machinery, infrastructure, or essential physical processes.
Manufacturing and Industrial Enterprises
Manufacturers rely on robotic systems, production lines, PLCs, industrial networks, and automated processes. A cyber incident can affect product quality, operational availability, worker safety, and intellectual property.
COE Security can help manufacturers assess industrial network security, evaluate IT and OT segmentation, identify vulnerabilities, conduct authorized penetration testing, review remote access controls, and strengthen incident response readiness.
Energy and Utilities
Electricity, gas, and utility operators depend on control systems to monitor and manage essential services. Security incidents may threaten availability, reliability, and public safety.
COE Security can help evaluate network architecture, access controls, exposed services, vulnerability management, monitoring capabilities, and recovery procedures to strengthen protection of critical operational environments.
Transportation and Logistics
Rail systems, ports, airports, warehouses, and logistics providers use connected operational systems to manage equipment, movement, scheduling, and physical processes.
COE Security can help assess network security, connected infrastructure, remote access, application security, third-party exposure, and incident response processes to support operational resilience.
Healthcare and Life Sciences
Hospitals, laboratories, and pharmaceutical manufacturers depend on connected medical, laboratory, building management, and production technologies. Disruptions can affect sensitive information and essential operations.
COE Security can help evaluate connected-system security, network segmentation, access management, vulnerability exposure, data protection, and compliance-focused security controls.
Government and Critical Infrastructure
Public-sector organizations and critical infrastructure operators may manage essential services, industrial facilities, and sensitive operational systems.
COE Security can help strengthen infrastructure security assessments, network protection, vulnerability management, threat monitoring, security governance, and incident response planning.
Technology and Industrial Automation Providers
Organizations developing industrial software, automation equipment, connected products, and IoT platforms must consider security throughout the product lifecycle.
COE Security can help conduct product and IoT penetration testing, review application and API security, assess cloud-connected components, evaluate secure development practices, and identify weaknesses in connected technology environments.
Conclusion
Booz Allen’s research highlights an important change in cybersecurity: AI agents may be able to navigate unfamiliar industrial environments and pursue complex objectives much faster than traditional assumptions about attacker effort would suggest.
The controlled laboratory results do not mean that every industrial facility is immediately vulnerable. They do show why organizations should not depend on obscurity, isolated security controls, or slow manual response as their primary defenses.
The way forward is to make industrial environments harder to traverse, reduce implicit trust, monitor consequential changes, strengthen segmentation, and preserve independent safety mechanisms.
As AI capabilities continue to advance, OT cybersecurity must bring together digital security, engineering expertise, operational resilience, and physical safety. Organizations that test their defenses, prepare for rapid response, and build security into industrial operations will be better positioned to protect people, equipment, and essential services.
About COE Security
COE Security partners with organizations in financial services, healthcare, retail, manufacturing, and government to secure AI-powered systems and ensure compliance.
Our offerings include:
- AI-enhanced threat detection and real-time monitoring
- Data governance aligned with GDPR, HIPAA, and PCI DSS
- Secure model validation to guard against adversarial attacks
- Customized training to embed AI security best practices
- Penetration Testing (Mobile, Web, AI, Product, IoT, Network & Cloud)
- Secure Software Development Consulting (SSDLC)
- Customized CyberSecurity Services
In addition, COE Security helps organizations strengthen operational and industrial cybersecurity through OT and ICS security assessments, network segmentation reviews, vulnerability management, secure remote access assessments, IoT and product penetration testing, threat monitoring, incident response planning, and security architecture reviews.
For manufacturing and industrial organizations, we help assess industrial networks, connected equipment, engineering workstations, remote access pathways, and IT to OT connections to identify security gaps and improve operational resilience.
For energy and utility providers, we help evaluate network security, access controls, infrastructure exposure, monitoring capabilities, and recovery readiness to support the protection of essential services.
For transportation and logistics organizations, we help assess connected infrastructure, applications, network architecture, third-party access, and incident response procedures to reduce cyber risk across operational environments.
For healthcare and life sciences organizations, we help protect connected systems, sensitive information, laboratory environments, and critical applications through security assessments, data governance, vulnerability management, and compliance-focused reviews.
For government and critical infrastructure operators, we help strengthen infrastructure security, network protection, threat monitoring, vulnerability management, and incident response readiness.
For technology and industrial automation providers, we help secure connected products, IoT devices, APIs, cloud platforms, and software development processes through penetration testing, secure development consulting, and product security assessments.
COE Security also helps organizations evaluate emerging AI security risks, including AI-assisted threat scenarios, excessive agent permissions, insecure integrations, and the governance controls required for responsible AI adoption.
Our goal is to help organizations identify security gaps, reduce cyber risk, protect critical systems, strengthen resilience, and maintain compliance as AI and industrial technologies continue to evolve.
Follow COE Security on LinkedIn for ongoing insights into safe, compliant AI adoption, industrial cybersecurity, critical infrastructure protection, emerging cyber threats, and practical security strategies to stay updated and cyber safe.
Click to read our LinkedIn feature article