The Human-in-the-Loop Imperative: Cyber-Operational Resilience in the Age of Frontier AI

The next major cyber incident may not begin with a dramatic breach announcement. It may begin with a sensor reading that does not match physical reality, a vendor outage that interrupts production, an AI system that can no longer be trusted, or a vulnerability exploited faster than the organization can understand its operational impact. That is the new reality of cyber resilience in the age of frontier AI: the risk is not only that attackers are moving faster, but that enterprises may no longer understand their own dependencies quickly enough to respond.

Artificial intelligence is changing the cyber risk equation faster than most governance models can adapt. A June 2026 statement from the Five Eyes cybersecurity agencies warned that AI is rapidly transforming cyber risk, increasing the speed, scale, and sophistication of threats, and shrinking the window between vulnerability discovery and exploitation. The agencies also emphasized that cyber resilience is now integral to business continuity, market confidence, and long-term value. That framing is essential: frontier AI models are accelerating vulnerability discovery, exploit development, social engineering, autonomous reconnaissance, malware adaptation, incident tempo, and velocity to impact, while also offering defenders powerful new capabilities for detection, prioritization, secure development, threat intelligence, and response.

For boards and executive teams, the implication is clear: cyber resilience can no longer be delegated as a technical issue. It is central to operational continuity, market confidence, safety, regulatory performance, and long-term enterprise value. The organizations that succeed will not be those with the most tools or the most ambitious AI pilots. They will be those that understand how cyber compromise, physical disruption, operational technology failure, third-party dependency, and AI model degradation can cascade into material business consequences—and that prepare their people, business processes, and governance frameworks accordingly.

1. The AI Shift in Cyber Risk Is Already Here

The cyber risk landscape has entered a new phase. AI is no longer a speculative trend or a narrow tool for security automation. It is already influencing how adversaries identify weaknesses, scale attacks, tailor deception, and accelerate exploitation. Frontier models and agentic systems are compressing key stages of the attack lifecycle that previously required scarce expertise and time. Reconnaissance, vulnerability analysis, exploit adaptation, phishing, malware variation, and target research can now be supported at machine speed and at a scale human teams cannot match.

Cybersecurity has always been asymmetric: attackers need one viable path in, while defenders must protect every critical asset, workflow, and dependency continuously. AI does not eliminate that asymmetry; it may intensify it. Offensive workflows are often easier to compress than defensive workflows. An attacker can move quickly from discovery to exploitation; a defender must validate risk, understand business impact, coordinate across teams, patch safely, avoid operational disruption, communicate decisions, and sustain services under pressure—all while defending dynamic business operations, attack surfaces, and threat landscapes.

The practical effect is a shrinking window between vulnerability discovery and exploitation. Assumptions about patch cycles, exposure management, and incident response timelines are becoming less reliable. Systems once considered manageable because exploitation required time, skill, or persistence may become more exposed as AI-enabled tools reduce the expertise required to exploit weaknesses. Legacy systems, unsupported platforms, exposed services, weak identity practices, and fragmented asset visibility are no longer just technical debt. They are strategic liabilities.

Recent cases show that AI-enabled cyber risk is no longer theoretical. In 2024, U.K. engineering firm Arup disclosed that fraudsters used AI-generated audio and video deepfakes to impersonate senior executives on a video call, inducing an employee to transfer about $25 million. OpenAI has reported disrupting more than 40 networks since early 2024 that attempted to misuse AI for malicious cyber activity, scams, and influence operations, including actors using AI to improve phishing, malware tooling, reconnaissance, and operational efficiency. Google Cloud’s Mandiant special report on AI risk and resilience similarly observed threat actors moving from experimental use of generative AI toward operationalized AI-enabled activity, including multilingual phishing, code troubleshooting, adaptive tooling, and the early use of AI agents and model-enabled workflows. These examples matter because they show AI entering the attack lifecycle through social engineering, reconnaissance, malware development, and enterprise trust relationships—not as science fiction, but as a practical amplifier of familiar tactics.

Yet the same capabilities that increase risk also create new defensive opportunities. AI can help security teams detect vulnerabilities earlier, correlate threat intelligence more effectively, reduce alert fatigue, improve software quality, model attack paths, prioritize patching, and respond more quickly to incidents. The question is not whether AI will be used in cyber defense. It already is. The more compelling question is whether organizations can integrate AI into governance, operations, and resilience practices in a way that strengthens the enterprise rather than adding new complexity, new dependencies, and new failure modes.

Crucially, organizations that are critical to national economies and the critical infrastructure that advanced nations depend upon have already experienced severe cyber events and some “near-misses” prior to the introduction of AI-driven attacks. The Jaguar Land Rover cyberattack was estimated to have cost the U.K. economy about £1.9 billion, or roughly $2.5 billion, and affected more than 5,000 organizations across the company’s supply chain. The Colonial Pipeline ransomware incident forced a proactive shutdown of pipeline operations and triggered a whole-of-government response to mitigate fuel-supply impacts. The ICBC Financial Services ransomware attack disrupted U.S. Treasury trade settlement, forced manual workarounds, and required a $9 billion capital injection to cover unsettled trades—the type of financial shock that triggers market meltdowns. These were not merely corporate incidents. They were warnings that cyber events can produce macroeconomic, energy-security, and financial-market consequences—systemic cyber events—and occurred prior to the accelerated velocity of potential damage now possible from AI further compressing attack timelines and increasing the risk of catastrophic business disruption.  

2. Cyber Resilience Is Now an Operational Leadership Issue

For too long, cyber risk has been described through the language of controls, vulnerabilities, incidents, maturity scores, and compliance obligations. Those remain important. But they do not, by themselves, answer the questions boards and executives must now ask with more urgency than ever: Which business processes fail if a key identity system is unavailable? Which operational technology environments can continue safely if visibility is degraded? Which third-party platforms are essential to revenue, logistics, customer service, or safety? Which AI-enabled capabilities depend on fragile data pipelines, models, training sets, compute environments, cooling systems, cloud services, or telecommunications links?

These are questions of business continuity, operational resilience, risk appetite, capital allocation, crisis governance, and strategic trust. A cyber incident that begins as a compromise of an IT system can halt industrial operations. A ransomware event can force executives to choose between safety, service continuity, regulatory obligations, and customer commitments. A disruption to a data center can impair AI-enabled analytics, fraud detection, logistics optimization, or cyber defense workflows.

The most consequential target in this environment is often not data. It is operational continuity. Operational technology, industrial control systems, supervisory control and data acquisition environments, building management systems, connected sensors, automation platforms, and field devices now form the nervous system of modern business. If these systems are manipulated, degraded, isolated, or made unavailable, the impact can move quickly from cyber incident to operational disruption, safety risk, financial loss, regulatory exposure, and reputational damage.

Cyber resilience therefore requires a shift from cybersecurity governance to resilience governance. Traditional cybersecurity governance asks whether controls exist. Resilience governance asks whether those controls will perform under pressure, whether leaders know what decisions they must make, whether teams can coordinate across functions, whether critical business services can continue when digital and physical systems are stressed simultaneously, how services will degrade gracefully, and how an organization will recover its critical business services if interrupted.

3. The Hidden Dependency Problem

Modern enterprise risk is no longer linear. A single business process may depend on cloud infrastructure, enterprise identity, operational technology, physical access systems, telecommunications, data centers, vendors, software libraries, industrial control components, and AI-enabled analytics. Few of these dependencies are visible in a traditional board dashboard, and a traditional dashboard is no longer sufficient.

This is the missing layer in many cyber risk programs: causality. Cyber-physical security is often discussed as the convergence of cyber and physical threats. That framing is useful but incomplete. The more important issue is how one domain causes disruption in another. How can a physical event create a cyber consequence? How can a cyber event create a physical or operational consequence? How can a third-party outage create an AI failure? How can a compromise of enterprise identity disrupt operational technology access, recovery sequencing, and incident command?

The industry has many tools for inventorying assets and measuring control maturity, but they leave decision support gaps that matter. During a fast-moving incident, leaders do not need abstract risk categories. They need decision-grade answers: what is affected, what could be affected next, what is the business impact, what can be isolated, what can continue safely, what must be restored first, and what trade-offs must be made now? Boards and executive teams should therefore demand a causal view of resilience.

4. Why Humans Matter More in an AI-Accelerated Environment

The rapid rise of AI has created a paradox: as automation becomes more powerful, human judgment becomes more important. This is especially true in cyber resilience, where consequential decisions often involve ambiguity, incomplete information, operational trade-offs, legal obligations, safety considerations, and reputational consequences. AI can analyze, correlate, recommend, and accelerate. But it cannot fully understand an enterprise’s mission, tolerance for disruption, regulatory obligations, customer commitments, or strategic context in the way accountable leaders must.

History offers useful analogies. During the Cold War, several nuclear “close-calls” were averted not because systems worked perfectly, but because humans exercised judgment under uncertainty. Vasily Arkhipov refused to authorize the launch of a nuclear torpedo during the Cuban Missile Crisis. Stanislav Petrov judged a Soviet early-warning alert to be a false alarm. Leonard Perroots later exercised restraint during the Able Archer 83 crisis, declining to recommend reciprocal escalation in response to signs of Soviet concern.

The lesson is not that humans are infallible. They are not. The lesson is that in high-consequence systems, resilience depends on informed human judgment, contextual interpretation, skepticism, and restraint. Automated systems can produce signals. Human leaders must decide what those signals mean, what action is justified, and what consequences may follow.

The same principle applies to cyber-operational resilience. Human operators often understand industrial systems, production processes, facility behavior, and operational anomalies better than any centralized monitoring tool. In some cyber incidents, frontline operators are the first to notice that equipment is behaving abnormally, that sensor readings do not match physical reality, or that manual procedures are no longer aligned with system outputs. Their knowledge can provide early warning, reduce false assumptions, and prevent automated responses from compounding harm.

At the board and executive level, the human-in-the-loop challenge is different but equally important. Senior leaders must be able to translate cyber signals into business decisions. They must understand when to isolate systems, slow operations, activate contingency procedures, notify regulators, engage law enforcement, communicate with customers, preserve evidence, or shift to manual operations. These decisions cannot be delegated entirely to AI systems or technical teams. They require authority, accountability, and judgment.

The Arup deepfake fraud is a useful warning for boards because it shows that human-in-the-loop does not mean “one human approves the transaction.” It means human judgment must be supported by resilient process design through strong governance: independent verification channels, out-of-band confirmation for high-risk actions, transaction thresholds, executive authentication procedures, escalation protocols, and a culture that gives employees permission to pause even when a request appears to come from the top. In an AI-enabled environment, trust must be designed, not assumed.

5. Frontier AI and the Boardroom Readiness Gap

The uncomfortable reality is that many boards are still early in the journey of understanding cyber risk as business risk. Many receive cyber reporting that focuses on technical metrics: vulnerabilities closed, phishing rates, control maturity, compliance status, endpoint coverage, or incident counts. These metrics are useful, but they often fail to explain how a cyber event would disrupt revenue, safety, production, service delivery, regulatory obligations, or market confidence. Boards need to know that the organization understands its assets and the critical business services dependent on them. Too often, this is a gap.

Frontier AI exposes this readiness gap. If AI-driven attacks compress the time between vulnerability discovery and exploitation, boards cannot rely on quarterly reporting cycles, static risk registers, or generic maturity updates. They need a dynamic understanding of exposure, dependency, and operational consequence. They need to know where the enterprise is brittle, which systems are essential to continuity, where manual workarounds exist, which third parties introduce systemic risk, and how AI-enabled capabilities depend on fragile infrastructure.

The board’s role is not to become technical. It is to ensure that cyber risk is governed with the same seriousness as other operational risks, financial risk, compliance risk, reputational risk, and strategic risk. Directors should expect management to provide clear answers to several questions: Which business services are most critical? Which assets and dependencies support them? Which cyber, physical, third-party, and AI scenarios could disrupt them? What level of disruption is tolerable? How quickly can the organization contain, continue, and recover? Which investments would reduce material business impact most effectively?

This requires the CISO’s role evolving from technical reporter to enterprise risk translator: a business leader focused on mitigating risk through effective security measures. In that capacity, the CISO and Chief Operations Resilience Officer, or a function comprising both roles, should work closely with the Chief Risk Officer to explain material risks (with an emphasis on the intersection of cyber and physical risks), assess their business impact, and present treatment options to the board in clear enterprise terms. The best CISOs do not ask boards to manage cyber controls. They help boards ask better questions about operational resilience.

6. Cyber-Physical Resilience in Gray-Zone and Hybrid Conflict

The strategic environment reinforces the urgency. Gray-zone and hybrid conflict are designed to create strategic effects below the threshold of conventional war. They rely on ambiguity, deniability, coercion, disruption, speed, and uncertainty. For businesses, this is not an abstract geopolitical concern. The same methods used to pressure states can disrupt commercial operations, critical infrastructure, logistics networks, financial services, energy systems, manufacturing, telecommunications, healthcare, ports, and supply chains.

NATO’s own framing reinforces this point: its guidance on countering hybrid threats describes modern hybrid activity as faster, more intense, and amplified by technological change and global interconnectivity, while emphasizing that national resilience and coordinated response capabilities are essential to defending societies and critical infrastructure against ambiguous, cross-domain disruption.

Hybrid disruption rarely appears as a clean, single-vector incident. A cyber intrusion may coincide with a drone sighting, a vendor outage, a communications disruption, an access-control failure, misinformation, or physical pressure near a facility. A drone incident near a critical site may become more severe if surveillance feeds, radio communications, incident-management systems, or building controls are degraded at the same time.

This is the reality of compound disruption. Adversaries may not need a single catastrophic attack to create strategic effect. Repeated ambiguous disruptions can erode confidence, delay delivery, increase costs, strain leadership decision-making, and expose governance weaknesses. In this environment, resilience depends on cross-domain coordination among cybersecurity, operations, physical security, legal, communications, risk, procurement, engineering, and executive leadership.

Tabletop exercises must evolve accordingly. Organizations should move beyond conventional ransomware scenarios and test compound events: a drone near a facility, anomalous control-system behavior, degraded communications, conflicting public narratives, third-party service disruption, and uncertainty about whether the event is criminal, state-linked, accidental, or coordinated. The value of these exercises is not only technical validation. It is executive rehearsal. They reveal whether leaders understand the business consequences of disruption, whether escalation thresholds are clear, and whether teams can make decisions under uncertainty.

7. AI as Both Defender and Dependency

Organizations should use AI deliberately to strengthen defense, but they must also recognize that AI itself becomes an enterprise dependency. AI-enabled security tools can improve vulnerability prioritization, anomaly detection, threat intelligence synthesis, secure code review, and incident response. AI can help defenders operate at greater speed and scale, particularly when security teams face overwhelming volumes of data and alerts.

But AI also introduces new resilience dependencies. AI-enabled business processes rely on data quality, model availability, compute, power, cooling, identity, cloud platforms, telecommunications, model governance, and secure integration. A cyberattack on an AI platform, a data poisoning event, a model supply-chain compromise, model drift obfuscation, a physical disruption to a data center, or loss of trusted data can become a business continuity event. Organizations that govern AI only as an innovation issue, model risk issue, or compliance issue will miss the operational dependencies that determine whether AI-enabled capabilities can function under stress.

The responsible path is not to slow defensive adoption out of fear. It is to adopt AI with resilience engineering in mind. Security teams should evaluate AI systems as tools, targets, and dependencies. They should ask how AI improves defensive workflows, how AI systems could be attacked or manipulated, and what business processes would be affected if AI-enabled capabilities were unavailable or untrusted. They should also ensure that human oversight remains meaningful, especially for high-consequence decisions involving containment, isolation, recovery, communications, legal obligations, and operational safety.

The NIST Artificial Intelligence Risk Management Framework provides a useful governance structure for this work because it encourages organizations to govern, map, measure, and manage AI risks across the AI lifecycle, reinforcing the need to treat AI systems as operational assets with resilience, security, accountability, and oversight requirements.

NATO’s revised AI strategy is also directly relevant because it treats AI as a general-purpose technology with distinct governance, security, interoperability, and adversarial-use risks, including issues related to generative AI, compute demands, human-machine teaming, and the need to protect and monitor AI technologies while preserving the ability to innovate responsibly.

8. The Practical Agenda for Leaders

The agenda for leaders is urgent but achievable. Foundational security practices matter more in an AI-accelerated environment because weaknesses can be discovered and exploited faster. Reducing unnecessary exposure, accelerating patching, addressing legacy systems, strengthening identity and access controls, and preparing for incidents before they happen remain essential. Secure-by-design and secure-by-default practices should become standard expectations, not aspirations.

However, resilience requires more than foundational hygiene. Leaders should build a decision-grade operating model that connects cyber risk to business impact. That model should include three priorities at a minimum.

1. Strengthen governance and decision rights. Define escalation thresholds, authority for isolation or shutdown decisions, board notification requirements, regulatory and customer communication triggers, and recovery priorities.

2. Use AI deliberately for defense. Apply AI to improve vulnerability management, threat detection, secure development, threat intelligence, and incident response while maintaining human oversight for consequential decisions.

3. Test under realistic compound conditions. Exercise scenarios that combine cyber, physical, OT, vendor, communications, and AI disruptions so leaders can practice decision-making before a crisis.

The central measure of success should be whether the organization can continue operating safely and credibly under pressure. Compliance matters, but resilience is proven in performance. A mature organization should be able to explain which critical operations can continue in degraded mode, how long they can be sustained, what decisions must be made, who has authority to make them, and what investments would most reduce material business impact.

9. From Cybersecurity Function to Strategic Capability

The organizations that lead in the next decade will transform cybersecurity from a defensive function into a strategic resilience capability. That transformation requires shared accountability. Cybersecurity teams cannot secure operational environments without operational context. Operations teams cannot manage cyber-physical disruption without cyber intelligence and response support. Physical security teams cannot address facility-level threats without integration into enterprise risk governance. Boards cannot oversee risk they do not understand in business terms.

Effective organizations will create a shared operating picture across cyber, physical, operational, legal, communications, safety, risk, procurement, and executive teams. They will connect resilience metrics to production, safety, service delivery, financial exposure, customer trust, and regulatory obligations. They will invest in people as well as technology, recognizing that skilled operators, informed executives, and practiced crisis teams are essential to resilience. They will treat AI as a force multiplier, not a substitute for accountability.

External collaboration will also become more important. No organization can fully understand AI-driven cyber risk, hybrid threats, supply-chain exposure, or sector-level targeting in isolation. The Five Eyes statement is notable because it frames the issue as a whole-of-organization and whole-of-society challenge, urging leaders to understand risk and accountability, prioritize foundational controls, empower cyber leaders, and stay actively engaged as threats evolve. Collaboration with peers, sector information-sharing organizations, government agencies, regulators, law enforcement, vendors, and external advisors strengthens situational awareness and response readiness. These relationships help leaders distinguish isolated incidents from broader campaigns, validate indicators, understand sector-specific risk, and coordinate response during disruption.

Conclusion: Act Before the Window Closes

Frontier AI is accelerating cyber risk while enterprises are becoming more dependent on complex digital, physical, operational, and AI-enabled systems. That combination creates a new leadership imperative. The risk is not only that attackers will move faster; it is that AI can make deception more convincing, reconnaissance more continuous, vulnerability exploitation more scalable, malware more adaptive, and social engineering more personalized. It is also that organizations may fail to understand how their own AI systems, data pipelines, models, cloud platforms, identity layers, and physical infrastructure create new points of operational fragility.

Cyber resilience in this era will not be achieved by automation alone. AI can accelerate detection, triage, analysis, and response, but it can also generate false confidence, amplify bad data, obscure accountability, and move faster than governance processes can validate. Resilience will come from the disciplined integration of AI-enabled defense, human judgment, operational knowledge, board-level governance, and cross-functional preparedness. Humans in the loop are not a sign of inefficiency. They are the mechanism through which organizations challenge machine outputs, interpret ambiguous signals, make accountable decisions, and preserve trust when systems are under pressure.

The path forward is clear: get the fundamentals right, govern AI as both a defensive capability and a resilience risk, rehearse compound disruption, empower cyber and operational leaders, and ensure boards can govern cyber risk as business risk. The organizations that act now will reduce exposure, strengthen operational continuity, and build confidence with customers, partners, regulators, and investors. Those that delay will face growing and avoidable disadvantage in a threat environment where AI is moving cyber risk from years to months—and from months to moments.

REFERENCES

• Cybersecurity and Infrastructure Security Agency. “Five Eyes Cyber Security Agencies Statement: The AI Shift in Cyber Risk—Why Leaders Must Act Now.” CISA, 22 June 2026.

• Tabassi, Elham. Artificial Intelligence Risk Management Framework (AI RMF 1.0). National Institute of Standards and Technology, Jan. 2023.

• North Atlantic Treaty Organization. “Summary of NATO’s Revised Artificial Intelligence (AI) Strategy.” NATO, 10 July 2024.

• North Atlantic Treaty Organization. “Countering Hybrid Threats.” NATO, updated 29 Jan. 2026.

• North Atlantic Treaty Organization. “Cyber Defence.” NATO, updated 30 July 2024.

• Elliott, David. “‘This Happens More Frequently Than People Realize’: Arup Chief on the Lessons Learned from a $25m Deepfake Crime.” World Economic Forum, 4 Feb. 2025.

• Nimmo, Ben, et al. “Disrupting Malicious Uses of AI: October 2025.” OpenAI, 7 Oct. 2025.

• Google Cloud. “AI Risk and Resilience: A Mandiant Special Report.” Google Cloud, 9 Mar. 2026.

• Pearson, James. “Jaguar Land Rover Hack Cost UK Economy an Estimated $2.5 Billion, Report Says.” Reuters, 22 Oct. 2025.

• U.S. Department of Energy, Office of Cybersecurity, Energy Security, and Emergency Response. “Colonial Pipeline Cyber Incident.” U.S. Department of Energy, May 2021.

• Kharpal, Arjun. “ICBC, the World’s Biggest Bank, Hit by Ransomware Cyberattack.” CNBC, 10 Nov. 2023.

• Dugar, Urvi. “Ransomware Attack on China’s ICBC Disrupts US Treasury Trades—FT.” Reuters, 9 Nov. 2023.

• Pearson, James. “Gang Says ICBC Paid Ransom over Hack That Disrupted US Treasury Market.” Reuters, 13 Nov. 2023.

• Center for Arms Control and Non-Proliferation. “The Soviet False Alarm Incident and Able Archer 83.” Center for Arms Control and Non-Proliferation, 14 Oct. 2022.

• National Security Archive. “Able Archer War Scare ‘Potentially Disastrous.’” National Security Archive, 17 Feb. 2021.