Omdia's Theresa Lanowitz reports a measurable shift: organizations are accelerating investment in offensive security capabilities, specifically penetration testing and red team tools powered by agentic AI. This isn't about adding another scanner to your stack. Security teams are fundamentally changing how they find vulnerabilities before attackers do.
The pattern is clear. As AI capabilities mature, your offensive security budget is moving from quarterly pen tests to continuous, AI-assisted attack simulation. But here's what the investment surge doesn't tell you: every capability you give your red team, you're also validating for potential attackers.
What the Data Shows
Three findings matter for your security program:
Investment in agentic AI for penetration testing is accelerating. Organizations are moving beyond traditional security testing cycles. Lanowitz's analysis points to teams deploying AI agents that can execute multi-step attack chains, adapt to defensive responses, and document findings without human intervention at every step. This represents a shift from quarterly external pen tests to continuous internal attack simulation.
The same AI capabilities enable both defense and offense. An AI agent that can identify a SQL injection vector, chain it with a privilege escalation, and exfiltrate data is demonstrating exactly what an attacker's AI could do. Your red team's new autonomous testing tool is a preview of next year's threat landscape. The gap between "what our tools can find" and "what attackers can exploit" is narrowing faster than most security programs can adapt.
Traditional pen test scoping breaks down with AI agents. When you contract a pen test, you define scope: "Test these five applications, exclude production databases, limit to business hours." AI agents don't naturally respect those boundaries. They explore laterally, test adjacent systems, and follow any promising path. Your existing rules of engagement were written for human testers who get tired and bill by the hour.
What This Means for Your Team
You're facing a practical problem: how do you adopt offensive AI capabilities without creating new attack vectors?
Start with your current pen test contracts. Most specify deliverables like "executive summary, technical findings, remediation roadmap." These contracts assume human testers who document as they work. AI agents generate findings continuously. You need new contractual language around data handling, scope enforcement, and what happens when an agent discovers something outside the original scope.
Your red team's AI tools need the same security controls as your production systems. If an AI agent can autonomously test your applications, it needs credential management, network segmentation, and audit logging. Treat your offensive security infrastructure as a high-value target, because it is. An attacker who compromises your red team's AI agent gains a pre-configured tool for mapping your environment.
The compliance implications aren't obvious until you hit them. PCI DSS v4.0.1 Requirement 11.3.1 requires external penetration testing at least annually and after significant changes. But "penetration testing" in that requirement assumes human-led testing with defined start and end dates. If you're running continuous AI-assisted testing, document how that satisfies the intent of the requirement. Your QSA will ask.
Action Items by Priority
Immediate (this quarter):
Review your current pen test contracts and rules of engagement. Add explicit language about AI tool usage, data retention, and scope boundaries. If your vendor is using AI agents, you need to know what guardrails they've implemented. Ask specifically: "How do you prevent your AI from testing out-of-scope systems?" and "Where is the data from our tests stored and for how long?"
Inventory any AI-powered security tools your team is already using. This includes GitHub Copilot for security code reviews, AI-assisted vulnerability scanners, and automated exploit frameworks. Map each tool to a specific use case and owner. You can't secure what you haven't inventoried.
Near-term (next two quarters):
Build a threat model for your offensive security infrastructure. Start with this question: "If an attacker compromised our red team's primary testing system, what could they access?" Include credential stores, network access, and any production system connections. Then apply the same controls you'd use for any high-privilege system: MFA, network segmentation, comprehensive logging.
Establish a review process for AI-generated security findings. AI agents excel at finding technical vulnerabilities but struggle with business context. A SQL injection in a deprecated internal tool requires different urgency than the same vulnerability in your payment processing flow. Human review adds the context that determines actual risk.
Create an AI usage policy for security testing. Define what AI tools are approved, what data they can access, and what human oversight is required. This isn't bureaucracy; it's risk management. Your policy should answer: "Can we use AI to test production systems?" and "What happens if an AI agent causes an outage?"
Strategic (annual planning):
Budget for continuous offensive security capabilities. The traditional model of annual or quarterly pen tests doesn't match the pace of deployment in modern development. Plan for tools and staffing that support ongoing attack simulation. This might mean hiring engineers with both red team and AI/ML skills, or contracting with vendors who offer continuous testing services.
Integrate AI-assisted testing into your NIST Cybersecurity Framework (CSF) implementation. Map offensive AI capabilities to CSF functions: Identify (asset discovery), Protect (validation of controls), Detect (testing of monitoring), Respond (tabletop exercises), Recover (testing of backup restoration). This mapping helps you articulate the value of offensive security investment to leadership.
