Let me begin with a story.
A few days ago, I was in a boardroom. One of those glass-walled, oak-paneled rooms where the air conditioning hums louder during silences. The CISO had just finished a 40-slide presentation on “security posture.” There were heat maps. There were NIST maturity scores. There were compliance percentages that would make any auditor weep with joy.
And then a board member, a quiet woman who had said nothing for the first 35 minutes, leaned forward and asked a single question:
“If we get breached tonight, what percentage of this company will still be making money tomorrow morning?”
Silence.
The CISO did not answer. Not because he was incompetent, but because his operational experience never taught him that this was the only question that mattered to the business.
For decades, cybersecurity operated on one core assumption: defenders had time. Time to detect, time to patch, time to investigate, and time to respond. We built our defenses like medieval castles with strong perimeters, high walls, limited trust, and the assumption that external threats could be kept out.
Since April 2026, Frontier AI systems have been able to continuously discover vulnerabilities at machine speed. Research consistently shows that AI-assisted attackers are compressing attack timelines from months into hours.
Microsegmentation and Identity are Foundational for Zero Trust
AI Is Disrupting How Enterprises Have Viewed Cyber Defense
Cyberattacks now occur at machine speed, with higher accuracy and the ability to weave together multiple data sources, resulting in far more complex attacks than ever witnessed. Overwhelmed defenders are now turning to AI to combat AI-powered cyberattacks, yet three key challenges remain.
Access the Best Practices Guide: AI Threat Resilience in the Age of Mythos
First: The AI-Powered Autonomous Attacks Have Just Begun
In July 2026, Hugging Face, the AI and machine learning platform, disclosed a high-severity data breach, marking a significant milestone in cybersecurity because it involved autonomous AI attackers. And OpenAI said that it lost control of two AI systems during a security test. The systems went rogue and hacked into the online startup Hugging Face. And the in-house AI system refused to use defensive maneuvers due to guardrails, compelling Hugging Face to use open-source AI tools to combat the situation.
OpenAI called the incident “unprecedented” but also predicted that such attacks would become increasingly common as AI adoption continues to grow.
Second: Poisoned Wells and Pure Springs
The 2026 SANS Institute Survey revealed a significant gap in governance policies designed to support the use of AI in cybersecurity, leading to more adoption failures. More than six in 10 practitioners said they have no visibility into where AI models are being used or what information is being exposed.
It is unprecedented that the same tool is being used for both attack and defense, making ingenuity the only difference between the attacker and the attacked. AI is no longer a tool, but it is a battleground. In today’s world, defenders cannot blindly trust automated shields.
Shields need to be untouchable by AI. For example, every AI-powered cybersecurity tool in the path of an AI can be exploited by an AI agent designed to find and exploit zero-day vulnerabilities. But if the AI agent cannot find a configurable asset, it is limited by preconfigured microsegmentation tools that reduce lateral movement.
Third: Higher Cybersecurity Investment Is Not Reducing Cyberattacks
Reuters reported that a large variety of US companies are facing a surge in AI-powered cyberattacks and ransomware that steal sensitive data and disrupt operations. These include:
- Consumer brands (Nike, Panera Bread, Hasbro)
- Tech firms (OpenAI, Crunchbase, Crunchyroll, Take-Two, Hugging Face)
- Healthcare/medtech (Stryker, Abbott, Clover Health, West Pharmaceutical, AdaptHealth)
- Critical infrastructure/aviation (Boeing)
Most of these are industry leaders, and they have multimillion-dollar investments in cybersecurity technology and large cybersecurity teams managing digital and AI initiatives. Obviously, we are missing the point.
AI-powered attacks can only be stopped by architecting digital navigation that does not allow lateral movement by design and reduces the attack options of an AI adversary. No path, no attack proliferation.
Read More: Microsegmentation: The Only Way to Stop Lateral Movement
AI excels at exploiting connectivity:
- Discovering reachable systems
- Chaining vulnerabilities
- Using stolen credentials
- Finding trust relationships
Microsegmentation reduces available connectivity. It does not try to outsmart the AI; it reduces its options.
We need to approach the challenge of AI-powered cyberattacks in a structured manner that makes sense to the board, the stakeholders, the CISO, and the rest of the digital organization.
Call to Action: Preparing for the Next AI-Powered Cyberattack
If you have read this far, you understand that you need to be breach-ready immediately. You need two plans. One tactical and the other strategic.
The Tactical Plan
- Implement microsegmentation for your digital infrastructure and lock down lateral movement. The fastest way to do that is to leverage your existing endpoint detection and response (EDR) investments (CrowdStrike, Microsoft, SentinelOne, Cortex, Trend Micro, etc.) and integrate them into your microsegmentation tool. Then implement modern cryptographic passwordless identity tools that can also govern access and authorities to ensure AI agents’ actions are controlled.
Access the Guide: The CISO’s Guide to Containment in the Age of AI Attacks
- Determine your critical digital infrastructure and divide it into zones and microsegments using controlled conduits. Struggling with zoning for breach readiness? Go by what you believe is your best material impact approach. Not by production/nonproduction environments, locations, or departments alone, but by which parts of your enterprise cannot be shut down, where you allow privileged access from, who your suppliers are, which systems you would want to allow AI agents to work within, etc.
- Enforce these policies and build playbooks for non-technical teams, especially leaders (CXOs), to communicate effectively internally and externally during unprecedented breaches. Gain executive delegation of authority where necessary and exercise these playbooks. This would give the board and other stakeholders confidence in the organization’s capability to prepare for the next breach.
The Strategic Plan
- Seek acceptable levels of material impact from your leadership. Assign those values to your assets and enrich your zoning and microsegmentation program. This will ensure that your digital enterprise, across data centers, operational technology (OT) systems, and the cloud, is architected for breach readiness.
- Establish the minimum viable digital enterprise (MVDE) that you target to keep operational even when AI-powered cyberattacks occur. Your microsegmentation rules must provide the governance capability to understand which digital systems might be breached and which will be significantly more difficult to breach. Establish immutable, tamper-evident audit logs for the entire MVDE.
Read More: Breach Readiness for the Post-Mythos AI Cyberattack Era
- Invest in tools that can go beyond simple prompt filtering to establish and govern AI guardrails at an application layer in real time. These tools should be able to use code-based hooks to enforce non-negotiable rules, prevent the agent from “creatively” interpreting your instructions, detect drift, use a large language model (LLM) firewall for real-time input/output inspection, enforce strict execution rails to govern tool usage, and conduct continuous red-teaming to adapt to emerging jailbreak techniques.
The time to do this is today and now. If we continue to develop our technology without wisdom or prudence, our servant may prove to be our executioner.
I am now of the firm opinion that cybersecurity has entered a phase in which breach readiness cannot be an enterprise-only milestone; it must also be a societal goal.
Roll up your sleeves. Get into the weeds. Get tactical first. Architect breach readiness. Govern how it is helping innovation. And how it keeps up with change. It is the only way to be ready for the uncertainty of AI.
To discuss how your organization can limit lateral movement and build breach readiness for AI-powered attacks, contact us.
