perspective
Before a cyberattack, there is a moment when every signal has more than one explanation. A privileged account signs in before dawn, but an administrator could be fixing an outage. Security logs stop arriving, but perhaps a sensor has failed. An AI agent queries a system it rarely touches, but it might be carrying out a legitimate task. The danger may be real, but the evidence is incomplete. That is the fog leaders face: a disruption may be a failure, an attack, or both—and while the cause may be unclear, the decision on how to respond cannot wait.
We are in an era of unprecedented technological change, and the critical systems and infrastructure our nation depends on are at risk. Frontier AI models are becoming weapon systems, offensive cyber is being democratized, and speed is the new currency of cyber operations. Growing dependence on connected systems, years of technical debt, AI-enabled threats, and accelerated decision cycles are testing our ability to sustain operations. We have yet to see AI-enabled attacks cause system-wide disruption, but AI is already reshaping every aspect of business, mission, and operations. Emerging agentic attacks point to a future where defensive cyber operations must move faster and with less human intervention. The speed gap between attack, understanding, and response is the new attack surface.
Over the past few weeks, outage reports and service disruptions have impacted a diverse group of big tech companies, large banks, utilities, and government agencies. Their scope and causes vary, and there is no public evidence they were connected or caused by cyberattacks. At the same time, multiple high profile cyber-attacks have occurred in recent months. AI driven cyber-attacks against South Korean and Japanese institutions, the recent theft of over $6 billion from cryptocurrency exchange Binance by DPRK cyber actors, and attacks against industrial refrigeration systems globally and water utilities in the U.S. raise the stakes of understanding what the next disruption means. These events have illuminated how much people and organizations depend on technology, and how quickly and severely technical failures and cyberattacks can affect customers and services across a connected environment, especially when operational resilience is weak.
AI and cybersecurity teams are under tremendous pressure. They must keep today’s environments running securely while simultaneously modernizing them for agentic AI operations. AI-enabled vulnerability discovery creates a race: defenders must discover and mitigate vulnerabilities faster than attackers can successfully detect and exploit them, even as teams handle more findings and patches. Agentic attacks are changing how adversaries operate, while decades of technical debt and poor cyber hygiene leave systems and infrastructure fragile. AI-generated code can speed development but also spread defects. Faster patching and deployment can introduce new risk, and rising agentic traffic adds complexity to environments not designed for it.
The broader risk comes from systems that are more complex and interdependent. Organizations patch and deploy more frequently, adopt AI across the software lifecycle, and rely on more cloud providers, vendors, platforms, and automated tools. Those dependencies can amplify a flawed update, maintenance action, or software defect across the business, obscure single points of failure, and make recovery as difficult as prevention. IT and cybersecurity leaders should work with business and senior leaders to assess where disruptions could spread and whether critical operations can continue. The answer is not to stop innovating. It is to build resilience into the way we operate. Operational resilience is not the absence of failure. It is the ability to continue the mission, contain the blast radius, and recover safely while the environment is changing faster than before.
These priorities apply to commercial industry and government alike. The aim is a minimum viable company (MVC): the essential functions and services that must continue through an attack, bad update, or provider failure:
An asset inventory is not enough. Define essential services, the level at which they must continue, and what can temporarily stop. Maintain a living map of applications, data, identities, networks, cloud services, vendors, software pipelines, and people that support them. Identify concentration risks, single points of failure, and know what stops if a dependency fails; update the map as the organization changes. Rehearse manual processes, backup communications, and workarounds to prove essential services can continue while recovery is underway.
A bad update can do as much harm as an attack. Fast patching and frequent deployment are necessary, but speed without control creates risk. Important changes should be staged, observable, reversible, and assigned to an accountable owner. AI-generated code and configuration should carry clear provenance and receive appropriate testing. Vendors should meet the same safe-deployment expectations.
Protect identity systems, cloud consoles, software repositories, CI/CD pipelines, signing systems, and management tools. If they fail, operations and recovery may stop. Give them separate access paths, independent communications, and tested recovery; ensure backups do not share a provider, identity system, software defect, or update mechanism. Enforce zero trust and micro segmentation to contain AI-driven intrusions and automated misconfigurations. Govern AI models and automation with continuous adversarial testing, human oversight, and lifecycle checks for drift, poisoning, and logic errors before widespread disruption.
AI-enabled cyberattacks are accelerating, and enterprises need to accelerate preparation. Run AI-enabled adversary simulations against your own environments to identify breakpoints in controls, response workflows, and recovery. Automate critical defensive processes so teams can act at machine speed. Implement asymmetric approaches such as counter AI to degrade adversary perception, disrupt automated decision loops, and make machine-speed attacks less reliable.
Assume critical systems will eventually fail or be compromised. The goal is not to prevent every failure, but to make it survivable. Test failover, manual operations, degraded capacity, communications loss, dependency outages, and recovery under realistic conditions. Resilience means limiting the blast radius, maintaining essential services, and recovering safely when a system fails. Lastly, have a communications plan that proactively keeps your customers, internal team, shareholders, and board informed, even when you are operating in your MVC mode. Stay ahead of the narrative attacks by setting the narrative early and with frequent updates.
Critical mission networks and processes are no longer just targets of traditional, human-paced intrusions; they now face machine-speed, AI-enabled attacks alongside the cascading risks of internal AI misconfigurations and automated system failures. Building deep, systemic operational resilience is no longer an optional technical upgrade—it is an existential imperative for mission continuity.
Ultimately, true operational resilience assumes that breach, disruption, and error are inevitable. By embedding zero-trust principles, designing for degraded operations, and enforcing disciplined AI governance, organizations can ensure that their most vital missions survive, adapt, and continue to function—even under the pressure of sophisticated cyber disruption. Our five priorities lift the fog and reduce the friction of AI by showing leaders which dependencies matter, where disruption can spread, and how to act.
The cause of disruption may be unclear. But what keeps the mission running shouldn’t be.
This paper uses the term artificial intelligence (AI) to remain consistent with cited research and incident reporting. Executive Order 14434 directs U.S. federal executive agencies to use Super Intelligence (SI) in place of AI in official communications; under the order, SI covers the same technologies as AI.