Fabio Fratucello explains why identity is becoming the new cyber battleground, urging enterprises to secure AI agents with just-in-time access, unified visibility and AI-driven detection and response.
As organisations increasingly adopt cloud, AI and autonomous systems, identity has emerged as a critical battleground in cybersecurity. Attackers are increasingly using valid credentials to move through enterprise environments, while the rise of AI agents is creating a new class of non-human identities with access to sensitive resources. At the same time, shrinking attack timelines are making human-led detection and response increasingly difficult.
In this interview, Fabio Fratucello, Vice President and Field CTO Worldwide, CrowdStrike, discusses why identity has become an attractive entry point for attackers, how enterprises can secure AI agents through just-in-time and just-enough access, why accountability for autonomous systems still rests with humans, and how organisations can build a unified, cross-domain security architecture capable of responding at machine speed.
CIO&Leader: Adversaries are increasingly choosing to log in rather than break into systems, using genuine-looking employee credentials or non-human identities. What gaps in traditional perimeter and endpoint security are making identity an easier entry point for attackers?
Fabio Fratucello: Traditionally, adversaries targeted endpoints because that is where runtime exists and where there is user interaction. A person is typing on the keyboard, making the endpoint an ideal environment to target.
However, EDR technology has made the endpoint more difficult to compromise. As a result, adversaries have started looking for other opportunities where they can gain an advantage.
That is where identity comes in. Identity-driven attacks, malware-free attacks and compromised identities have been trends we have seen for several years. Adversaries are looking to obtain usernames and passwords because they can then log in rather than break in.
From a defensive perspective, logging in with a valid credential generates less noise. There are no brute-force attempts or authentication failures to immediately trigger defensive mechanisms. While behavioural signals may still raise concerns, detecting and responding to an attacker using a legitimate identity becomes much more challenging.
CIO&Leader: When attackers blend into normal operational traffic using valid credentials, how can security teams distinguish between legitimate activity and a compromised identity in real time?
Fabio Fratucello: Identity threat detection and identity threat protection capabilities can make a significant difference. Authentication is an important part of the puzzle, but it is only one signal.
Identity threat protection and identity threat detection are key capabilities from a detection and response perspective, but then again, this is one piece of the puzzle.
Logging into a laptop, for example, may simply be the first step in an attack. The adversary may then attempt to move laterally to another laptop, server or system, either on-premises or in the cloud. Organisations therefore need visibility into the entire sequence and the ability to intervene at that level.
The challenge is that breakout time is becoming shorter. Over the last five years, adversaries have continued to reduce the time required to move through an environment. The last Global Threat Report recorded an average breakout time of 29 minutes, while the fastest e-crime breakout time was less than 30 seconds.
For defenders, that means human-initiated response is increasingly unlikely to be fast enough.
Identity threat detection and protection therefore need to be part of a broader platform capable of understanding the entire attack. We call this a cross-domain attack, where an adversary moves from an endpoint to an identity and then to another logical or physical part of the organisation.
With such short response windows, AI capabilities are also becoming essential because defenders increasingly need machine response time to counter sophisticated adversaries.
CIO&Leader: As organisations deploy autonomous AI agents and non-human identities, what unique security risks do AI agents introduce compared with traditional service accounts or human employees?
Fabio Fratucello: This is one of the major shifts we are seeing. Organisations are embracing AI to deliver faster and better outcomes, and now we have agents that can effectively operate as superhuman employees.
Agents can initiate work and activity, access resources, and use identities to authenticate into different parts of an organisation. The major difference is that they operate 24/7, do not get tired and can process enormous amounts of information in a very short time.
We need AI that is designed to deliver defensive outcomes, and we need that to be part of the platform to be effective.
From a defensive standpoint, this makes identity controls even more important.
The fundamental capabilities applicable to human identities also apply to AI identities, but organisations need to go further. Concepts such as just-in-time and just-enough access become critical.
An agent should only be able to access particular resources for a defined period and only have the level of access necessary to perform its task. These controls create guardrails around what an agentic identity can do.
That becomes particularly important as we see emerging attacks such as direct and indirect prompt injection, model tampering and skill abuse. Strong identity controls can limit the potential impact of such attacks.
CIO&Leader: If AI agents can make decisions, access databases and delegate tasks independently, who ultimately owns accountability when an agent breaches a security boundary?
Fabio Fratucello: I don’t think there is a straight answer at this point because organisations are likely to handle accountability differently.
However, in the majority of organisations we work with, the person, team or human remains accountable to the C-level and the board. At this point, accountability is not something we have seen organisations being able to pass down to the machine.
That is why governance, visibility and guardrails are becoming central to AI adoption. Leaders need confidence that autonomous systems are operating within defined controls and within the organisation’s risk appetite.
The journey generally starts with visibility. Organisations first need to understand how AI is being used internally, including generative AI accessed through browsers and agents operating on endpoints.
Once visibility is established, organisations can introduce governance and guardrails, such as authorising specific services and restricting others.
More mature organisations are then moving towards AI Detection and Response, or AIDR. At that stage, organisations typically have an AI strategy, dedicated teams and a clearer understanding of their AI adoption. They can then introduce technical controls to protect agents from adversaries, human mistakes, data exfiltration and data leakage.
Ultimately, accountability continues to sit with the security leader, CIO or whoever represents the AI function at the leadership level.
CIO&Leader: Why have static access policies and standing privileges become a liability in the age of autonomous AI? What does continuous, risk-aware authorisation look like?
Fabio Fratucello: Organisations at the forefront have already moved from static privilege to just-in-time privilege, even for human users.
The principle is straightforward. A user may need a particular privilege to perform a specific activity, but once that activity is completed, there is no reason to maintain that level of access. Reducing privileges reduces the potential blast radius if the account is compromised.
The same principle becomes even more important with AI agents because they are effectively superhuman. They can perform activities at much greater volume and speed than a person.
Imagine an agent with domain credentials. If an adversary compromises that agent, they could potentially tamper with it, modify how it behaves and use those credentials to operate at the domain level.
That can become a very serious problem very quickly.
The ideal scenario is for the agent to have access only to limited resources and only the ability to read or modify what is necessary. If the agent is compromised, the attacker is similarly constrained.
That gives defenders more time to detect, respond and contain the incident. The more difficult and time-consuming we make the attack for the adversary, the greater the likelihood that they will make a mistake, generate noise and give defenders an opportunity to respond.
CIO&Leader: How can organisations bring human, non-human and AI identities under a unified security architecture without slowing down business operations?
Fabio Fratucello: The problem is becoming increasingly relevant because organisations are using more technologies, generating more data and creating more potential attack domains.
Looking at these domains individually is no longer sufficient. A security tool may provide a good picture of what is happening in one area, but that does not necessarily explain the entire attack or what is likely to happen next.
The best approach is a modern platform that provides cross-domain visibility and AI capabilities for defenders.
The objective should be a harmonised view rather than simply bringing different technologies into one console. Identity telemetry and endpoint telemetry, for example, need to speak the same language and be presented cohesively so defenders can understand the entire attack story.
It is like reading an entire book rather than reading only Chapter 3. If someone logs into a laptop and then begins moving laterally, understanding what happened before that movement provides important context for determining whether it is legitimate activity or part of an attack.
Speed is equally important. Adversaries are increasingly using automation and AI to accelerate their activities, while enterprises are generating terabytes and petabytes of data. Humans alone cannot effectively process all that information.
We therefore need AI designed to deliver defensive outcomes, and that capability needs to be integrated into the security platform itself.
CIO&Leader: As AI agents begin interacting with other AI agents across partner ecosystems, what should CISOs build today to ensure long-term identity governance?
Fabio Fratucello: The fundamental identity concepts remain similar even when interactions extend beyond a single organisation.
Businesses have long connected with third parties through SaaS platforms. They use identities to authenticate, exchange data and process information.
The medium is now changing. Instead of a person logging into a browser, an AI agent may use an MCP server to communicate with another AI agent. There may also be an AI gateway, API calls or other technical pathways through which information moves.
But the control capabilities remain broadly similar.
When CrowdStrike thinks about AIDR, we are therefore not looking only at compute. We also look at identity interactions across the broader ecosystem.
Organisations need to understand whether there will be MCP calls, API calls or AI gateways and determine what protection mechanisms apply at each level.
The principles of visibility, detection and response remain central. The technology pathways may change, but the need to understand identities, monitor their behaviour and respond when something goes wrong remains fundamentally the same.