ai-driven cybersecurity

How AI-driven Cybersecurity Is Transforming Risk Management

AI-driven cybersecurity: This interview explains how AI is changing cybersecurity, creating new challenges for vulnerability management, and what technology leaders need to do to stay ahead of emerging threats.

For years, cybersecurity teams have raced to discover vulnerabilities before attackers could exploit them. Artificial intelligence is changing that equation. Today, AI can help defenders identify threats faster, but it also enables attackers to discover weaknesses, automate exploits, and launch highly targeted attacks at unprecedented speed. As a result, the window between vulnerability discovery and exploitation is shrinking from weeks to days – and, in some cases, even minutes.

This shift is forcing organizations to rethink how AI-driven cybersecurity fits into their long-standing security practices. Today, simply finding more vulnerabilities is no longer enough. Security teams must also sift through a growing flood of AI-generated findings and respond before those weaknesses can be turned into real attacks.

To better understand this changing cybersecurity landscape, we spoke with Heath Mullins (Chief Evangelist at ExtraHop and Former Forrester Analyst). He shares his perspective on the challenges AI is creating for enterprises and offers practical advice on how technology leaders can prepare for machine-speed attacks.

Rethinking Vulnerability Management in the AI Era

Finding vulnerabilities is no longer the biggest challenge. The difficult part is knowing which ones matter most. With AI producing thousands of security alerts, organizations need smarter ways to prioritize risks and respond before attackers can take advantage.

In your opinion, what is a greater challenge: finding vulnerabilities or managing the overwhelming volume of AI-generated findings? How should security leaders rethink their vulnerability management programs in this new reality?

Mullins: Finding vulnerabilities and managing the volume of AI-generated findings are both notable challenges. Attackers have gotten more savvy and sophisticated, creating the challenge of detection, but in the same vein, AI powered alerts within the SOC have created a ton of noise for security analysts to deal with, in addition to regular proactive activities like threat hunting.

Today’s security teams cannot possibly prioritize every vulnerability. And instead of treating all “critical” CVEs as equal they must prioritize findings based on environmental context and business impact. In order to do this security professionals need to look for signals that indicate a critical detection, such as where vulnerabilities intersect with public reachability, sensitive data, and overly permissive identity access.

How does AI fundamentally change the economics of cyber offense and defense. And who stands to gain the advantage over the next five years?

Mullins: One of the biggest shifts I have seen brought on by AI is that attacker expertise is no longer a prerequisite for cybercrime. Individuals with little to no technical backgrounds can use tools to identify vulnerabilities and exposed assets and automate workflows that previously required years of experience. Now with one tool, less expertise and a lower cost, even the most amateur hackers can easily carry out large scale, devastating campaigns that boost hackers’ ROI.

One attack vector specifically that has been propelled by AI is social engineering. Attackers can now create highly personalized and targeted campaigns at scale. They are able to scrape digital footprints and craft highly convincing, context-aware phishing campaigns that bypass traditional spam and suspicion filters at half the cost of just 5-10 years ago. However, the same can be said for defenders who are deploying AI and fundamentally altering their cost structure. Strategic deployment of AI can reduce the average breach costs by up to $1.9 million and shorten breach lifecycles by approximately 80 days.

At the same time, AI-driven efficiencies are only one part of a broader shift reshaping the cybersecurity landscape. Defensive capabilities that were once available only to well-funded organizations are now becoming increasingly accessible to smaller firms.. The market is undergoing a structural change and beginning to prioritize recovery over prevention. Consumption-based models are making enterprise-grade cyber resilience more affordable for mid-market companies. With predictable monthly pricing, these models not only create recurring revenue opportunities for vendors. But also help SMBs defend against increasingly sophisticated cyberattacks.

Are organizations at risk of creating a false sense of security by relying too heavily on AI-powered vulnerability discovery tools without investing equally in remediation capabilities?

Mullins: There seems to be a widening gap between how much an organization uses and deploys AI and how ready their infrastructure is to handle the extra pressure. This rift can be dangerous for many reasons. But the primary concern is a false sense of security that this could create. Many IT professionals are starting to admit that AI adoption is moving faster than their ability to protect their organization against new threats. The challenge stems from trying to secure a dynamic, fast-moving AI layer on top of a rigid, siloed base.

To close this dangerous loophole, industry leaders suggest shifting focus from detection to containment and active remediation. Organizations must update risk models, improve cross-functional decision-making, and make sure they have the right data and context powering AI-powered tools.

Why Threat Intelligence Must Become More Collaborative

As AI-driven cybersecurity continues to evolve, organizations can no longer tackle threats alone. Staying ahead of attackers requires greater collaboration, timely threat intelligence sharing, and a more coordinated response.

Frontier AI companies such as Anthropic are already moving away from tightly controlled cybersecurity programs. Instead, they are embracing collaborative threat intelligence. What do you think about this shift?

Mullins: Frontier AI companies are actively pivoting from closed programs to a system of collaborative intelligence to counter AI-driven threats. But I think more needs to be done.

For example. Anthropic launched its information-sharing initiative, granting approximately 200 trusted partners in over 15 countries access to specialized AI models called Project Glasswing. One of those models, Mythos, can identify security vulnerabilities at unprecedented speed. By giving select partners early access, defenders have a chance to understand how these tools work and prepare before similar capabilities fall into the wrong hands.

These partnerships and projects are beneficial. They will create faster response times, strengthen critical infrastructure, and deepen technological visibility. However, more needs to be done when it comes to navigating the flood of AI-generated security findings quickly enough before attackers exploit them.

Part of this initiative should encourage organizations to focus on the vulnerabilities that pose the greatest risk. Patching efforts should be prioritized based on that level of risk. This is especially important as the Cyber Information Sharing Act is set to expire in September.

Could we reach a point where sharing AI-discovered vulnerabilities becomes a regulatory expectation rather than a voluntary practice?

Mullins: We are already moving toward a world in which AI-discovered vulnerabilities become a regulatory expectation. Regulators are starting to realize that keeping AI-discovered vulnerabilities hidden in voluntary silos leaves critical infrastructure disastrously exposed. CISA announced that they are issuing a new directive to improve how federal agencies prioritize mitigating cyber vulnerabilities. This directive is part of CISA’s response to the current threat landscape, where AI software services can assist threat actors to find and exploit vulnerabilities. They have outlined expectations for when and how to check if a vulnerable system was compromised by a threat actor before the patch was applied.

Shrinking Window Between Discovery and Exploitation

AI is dramatically reducing the time between vulnerability discovery and exploitation, leaving organizations with far less time to react. Success will depend on how quickly security teams can detect, prioritize, and respond before attackers gain the upper hand.

If defenders are drowning in AI-generated findings, attackers are likely using similar technologies. Are we entering an era where speed of response becomes more important than prevention itself?

Mullins: I wouldn’t necessarily say speed of response will become more important than prevention. But in order to have robust security resilience, cybersecurity teams definitely need to balance both. However, no matter how mature your cyber strategy and tactics are. It is still very likely that an attack is inevitable. With that mindset, the speed at which you can identify, respond, and mitigate an attack becomes extremely crucial to business continuity.

Like many security questions these days, AI is both the problem and the solution. As both attackers and defenders use AI to launch attacks and prevent evasion in their networks, AI speeds up and slows down the attack timeline. Cyber criminals are using this technology to quickly gain access to environments and rapidly exploit vulnerabilities. However, security defenders who have deployed AI to analyze and monitor network traffic may be drowning in AI alert fatigue. They may be unable to differentiate between real threat alerts and other flags.

In order for companies to strike an equilibrium, there are a couple of ways they can better sift through the notifications in the SOC. They should first always be working under the “assume breach mentality”. This allows them to operate under the assumption that adversaries are already inside the network. The priority then becomes detecting lateral movement quickly and containing it through micro-segmentation, rather than relying solely on perimeter defenses.

Secondly, organizations need a strong baseline of enterprise activity and data to support their AI-powered tools. This helps reduce noisy alerts and improves the accuracy of AI-driven security workflows.

How quickly do you expect threat actors to operationalize AI-discovered vulnerabilities once these technologies become widely accessible?

Mullins: The next generation of threats will adapt in real-time. Machine learning has dramatically compressed the exploitation timeline by eliminating the grace period between vulnerability disclosure and weaponization.

Cybercriminals and nation-state hackers are moving much faster to take advantage of software weaknesses. Some groups were able to turn newly announced software flaws into working attacks in only a couple of days. And with the surge in vulnerability discovery and sharing, that timeline will most likely shrink to minutes.

Could AI eventually create a situation where the window between vulnerability discovery and exploitation shrinks?

Mullins: The question isn’t if that could happen in the future. It’s what organizations are going to do when this inevitably happens sooner than most would think. For example. AI tools can now analyze vulnerability reports, search large codebases, propose fixes, generate tests, and explain remediation steps in minutes rather than hours or days, allowing human security teams to respond much faster.

The compression of the exploit timeline is driven by three main AI-driven capabilities. They are automated patch tracing, weaponization at scale, and autonomous scanning and reconnaissance. When vendors release security updates, AI models quickly compare the new and old code to pinpoint exactly what was fixed, revealing the underlying weakness. Agentic AI systems can also convert abstract vulnerability descriptions into repeatable, functioning exploit payloads. Once an exploit is generated, machine-speed models can sweep the entire internet to locate vulnerable, unpatched endpoints in hours, not weeks.

The difference, however, between attackers and enterprises is that attackers don’t have compliance, regulation and manual tests. Weaponized AI can exploit a vulnerability and deploy a scanner in minutes. While the average enterprise requires weeks to manually triage, safely test, and push those same security patches into production environments. In order for enterprises to survive in a world where attacks occur at machine speed, they need to implement a fundamental shift in how they can react to these high-velocity threats at speed and scale.

Building Security Strategies for the AI Era

Technology alone cannot solve modern cybersecurity challenges. Organizations also need the right metrics, governance models, and cross-functional processes to ensure AI improves security without sacrificing trust or compliance.

What new metrics should CISOs and CTOs track in an AI-driven security environment?

Mullins: Current AI measurement frameworks suffer from a fundamental capability-performance gap. They typically measure what the machine can do in a vacuum, rather than what it actually achieves when a breach is in progress. Today’s measurements are essentially “vanity metrics” for AI. They track the mechanics of labor rather than the integrity of defense.

Therefore, there are a handful of metrics that CISOs and other security executives should prioritize for AI-driven security environments. Such as mean time to respond (MTTR), mean time to detection (MTTD), and alert volume reductions. These metrics will help leaders identify how quickly AI models ingest, correlate, and enrich a raw alert with business context, reducing human investigation time. Improving alert volume reductions directly helps teams improve MTTD and MTTR. By filtering out the noise and false positives, analysts can investigate actual threats much faster, driving down response times.

What organizational changes are required to ensure security teams can make faster decisions without sacrificing accuracy and governance?

Mullins: The only way to enable faster security decisions without sacrificing accuracy or governance is to adopt a security-by-design approach. This requires establishing policy-driven guardrails and deeply embedding compliance requirements into daily workflows.

There are a handful of steps security teams can take in order to ensure they are prioritizing security while also using advanced technologies to make decisions. They should replace manual ticket-based reviews with automated “compliance-as-code” and integrate continuous vulnerability scanning and secure code checks. They should also move away from isolated, siloed security teams and establish a shared responsibility model where teams evaluate risks together.

This will ensure that security considerations are inherently built into product development and IT deployment.

What Tomorrow’s Cybersecurity Leaders Will Do Differently

The next generation of cybersecurity leaders will be defined not by the number of vulnerabilities they discover. But by how effectively they automate remediation and build resilient security operations. Leaders who embrace AI-driven cybersecurity strategically today will be better prepared for tomorrow’s threat landscape.

Looking ahead five years, what will distinguish cybersecurity leaders from laggards in an era where AI can identify millions of potential weaknesses across an enterprise environment?

Mullins: Five years from now, organizations that continue to manage vulnerabilities using manual human triage and legacy patch cycles will be in an endless game of catch-up, falling behind their competitors and losing their market advantage.

Leaders, on the other hand, will rely on AI-driven autonomous remediation that acts without human intervention for routine fixes. Instead of patching everything based on severity, they will prioritize fixes based on the likelihood of an active exploitation.

Additionally, companies implementing an agentic SOC today will face less alert fatigue as their agents and data mature, allowing these autonomous workers to independently identify, verify, and block malicious activity in real-time across the enterprise.

Advice

As AI-driven cybersecurity reshapes every aspect of enterprise architecture, technology leaders face difficult decisions about adoption, governance, and risk. Heath Mullins closes with his most important advice for organizations preparing for an AI-driven future.

If you could give every CTO and CISO one piece of advice as AI reshapes the cybersecurity landscape, what would it be – and why is acting now more critical than ever?

Mullins: Don’t get overwhelmed. Stay grounded in solid security principles, and thoroughly vet every AI instantiation within your network, shadow or otherwise. Security leaders need to embrace the concept of AI vs. AI, or risk becoming laggards and vulnerable to exploitation by less sophisticated attackers with access to bleeding-edge tools. Stay ahead of the curve if at all possible or at least have the tools in place to monitor your networks and infrastructure for signs of an attack at the very least.

Key takeaway:

AI is changing the speed and scale of cybersecurity on both sides of the battlefield. Organizations can no longer rely on traditional vulnerability management practices. Instead, they need a more contextual approach – one that focuses on business impact, exploitability, and rapid remediation.

For tech leaders, the message is clear: resilience will become the defining characteristic of successful cybersecurity strategies. Organizations that can quickly detect, prioritize, and respond to threats at machine speed will be better positioned to protect their businesses than those that continue to rely on manual processes and reactive security models.

About the Speaker: Heath Mullins is Chief Evangelist at ExtraHop, where he leads thought leadership and advocacy around advanced cybersecurity solutions. Before joining ExtraHop, he served as a Senior Analyst at Forrester, advising Global 100 enterprises, U.S. federal agencies, the Department of Defense, and allied governments on cybersecurity strategy. A recognized voice in the industry, Heath is known for his expertise in Zero Trust, cyber resilience, and threat detection for highly regulated organizations.

Gizel Gomes is a technical writer with over 16+ years of professional experience in technology journalism, content development, and editorial writing. Holding a bachelor's degree in Computer Science, she writes extensively on artificial intelligence, cybersecurity, cloud computing, data centers, enterprise software, digital transformation, and emerging technologies. She is also known for producing insightful executive interviews and thought leadership features that bridge the gap between technical innovation and business strategy, helping technology leaders make informed decisions in an evolving digital landscape.