AI hasn't changed what good security requires. It's changed what happens when you don't have it.
Breakout time, the window between initial access and an attacker moving laterally through a network, is now under 30 minutes in many incidents. Exploitation follows vulnerability disclosure within 24 hours. AI-generated spear phishing is hitting click-through rates of 50-60%, compared to single digits with traditional phishing lures. The attacks are faster, more targeted, and harder to distinguish from legitimate activity.
The same security fundamentals have always mattered. There's just now far less margin for error.
That's the actual definition of an AI-ready security program. Not a new framework. Not a new product category. A standard for how well the fundamentals have to work now, because the cost of gaps that were once recoverable has become catastrophic.
An AI-ready security program is one built to operate at the speed and scale of AI-accelerated threats. It reduces exploitable attack surface, detects across identity and endpoint, correlates signals across domains, baselines normal behavior, and can contain an active incident within minutes, not hours.
The five criteria below define what that looks like in practice.
The most effective security win available is not giving attackers something to target.
AI-accelerated reconnaissance is fast. Tools that once required hours of manual effort now sweep public infrastructure, enumerate exposed services, and build target profiles at scale. If you're running legacy VPN infrastructure with known CVEs, operating with over-permissioned accounts, or leaving cloud services exposed without identity-based controls, those gaps will be found.
Attack surface reduction used to be framed as a nice-to-have. It's now essential. The specific shift worth noting: traditional SSL VPNs carry a different risk profile than agent-based Secure Access Service Edge (SASE) solutions. A VPN appliance is a publicly reachable endpoint that can be directly exploited. SASE ties access to device identity and user identity simultaneously, removes the publicly addressable target, and enforces zero-trust access policies at the session level.
An AI-ready program has a documented answer to the question: if an attacker were targeting us today, where would they find their way in? If that answer is unclear or uncomfortable, that's the starting point.
85% of attacks observed in Todyl's Security Operations Center (SOC) operate entirely at the identity layer and never pivot to the endpoint. They come in through stolen credentials, compromised email accounts, or SSL VPN access using valid usernames and passwords. There is no malware. There is no novel exploit. The attacker logs in.
This has a direct implication for detection coverage: if your visibility starts at the endpoint, you are already behind on most incidents.
Identity detection must be more than what your identity provider offers natively. Microsoft and Azure's built-in detections are a floor, not a ceiling. An AI-ready security program builds its own anomaly framework on top, enriched with external threat intelligence to understand what IPs and sessions look like right now, not just what they looked like historically.
Impossible travel detection, behavioral baselining, OSINT enrichment on login IPs, and industry-contextualized alerting all belong in this layer. None of them require exotic capabilities. They require intentional architecture.
A single alert that fires in isolation is not detection. It's an input to detection. The difference matters enormously at AI-attack speeds.
Here's what that gap looks like in practice: a valid credential used from a new device triggers a low-confidence alert. An unusual PowerShell execution on a workstation triggers a separate low-confidence alert. An inbox forwarding rule gets set. Each event, reviewed in isolation, has an innocent explanation. Reviewed together, in sequence, they describe an active account takeover in progress.
Traditional MDR and siloed detection stacks don't see the story. They see three separate events in three separate dashboards. By the time a human analyst correlates them manually, the attacker has staged data and moved.
An AI-ready detection architecture treats every alert as the anchor for an investigation, not a conclusion. It automatically pulls related signals across identity, endpoint, network, and cloud. It asks: what else was this user doing? Does this match their pattern? What happened in this session before the alert fired? That context converts low-confidence noise into a high-confidence case, and it has to happen at machine speed, because attackers are no longer operating at human speed.
Attackers have known for years how to avoid signature-based detection. They use legitimate admin tools: Remote Monitoring and Management platforms, PowerShell, PsExec. They log in with real credentials. From a signature perspective, everything they do looks like normal IT activity.
The baseline question for any security program in 2026 is: what does normal look like for each user and device in your environment, and how quickly can you detect deviations from it?
User and Entity Behavior Analytics (UEBA) is one piece of this. If a user who has never run PowerShell suddenly runs a script to enumerate the network, that's not a signature violation. It's a behavioral one. If a credential authenticates from Denver at 9 am and Chicago at 10 am, that's not a malware signature. It's an impossible travel event.
The other piece is understanding the full session. An AI-ready program doesn't just ask whether a login location has been seen before. It asks whether this session looks like what this user does: are they opening their regular documents, emailing their regular contacts, following their regular workflow? That depth of session-level analysis is what closes the gap between catching an attacker on day one and finding them three weeks later.
The final criterion is response speed, and it's the one that determines whether an incident becomes a recoverable event or a business-ending one.
Network segmentation enforced through SASE and LAN Zero Trust (LZT) limits how far an attacker can move once they're inside. Even with valid credentials, zero trust constrains lateral movement to only the resources that identity is authorized to reach. In a world where breakout time can be under 30 minutes, that constraint buys critical time.
But segmentation alone isn't enough. The response function needs to match the threat function. Manual triage cycles that take hours to reach a conclusion are structurally too slow for AI-accelerated attacks. An AI-ready program has automated response playbooks, 24/7 coverage, and the ability to isolate, contain, and investigate across identity, endpoint, network, and cloud, not just endpoint.
The question worth asking is if an attacker gained access right now, how quickly could you contain them?
I close every conversation about AI-ready security with three diagnostic questions.
Where is your attack surface still open? If you were an attacker, where would you find your way in? Are you OK with that exposure?
Where are your detection coverage gaps? Can you see across identity, endpoint, network, cloud, and SaaS? Do you have the visibility to investigate incidents as they happen, not reconstruct them afterward?
How fast can you contain? When an incident happens, do you have the automated response, 24/7 coverage, and cross-domain response capability to stop it before the damage compounds?
These are diagnostic questions. An organization that can answer all three with confidence has built something that will hold up against AI-accelerated threats. An organization that can't has identified exactly where to start.
An AI-ready security program is one built to operate at the speed and scale of AI-accelerated threats. It reduces exploitable attack surface, detects across identity and endpoint, correlates signals across domains, baselines normal behavior, and can contain an active incident within minutes, not hours.
The five criteria, summarized:
According to the World Economic Forum, 94% of organizations identified AI as the dominant cybersecurity force shaping 2026. The bar for an attacker to conduct sophisticated tradecraft keeps getting lowered. Organizations must prepare for AI-accelerated attacks.
Check where your security program stands. Take our free AI-readiness assessment.
Traditional attacks often gave defenders hours or days to detect and respond. AI-accelerated attacks compress that window dramatically. Breakout time in many incidents is now under 30 minutes. Vulnerabilities are exploited within 24 hours of disclosure. AI-generated phishing achieves click-through rates of 50-60%. The fundamentals of good security haven't changed; the margin for error has.
Behavioral detection identifies threats by recognizing deviations from established patterns of normal activity, rather than matching known attack signatures. If a user who has never run PowerShell suddenly executes a network enumeration script, or a credential authenticates from two cities within an hour, behavioral detection surfaces those events as threats even when no known malware signature is present.
Breakout time is the window between an attacker's initial access to a network and their first lateral movement to another system or resource. It is one of the most important metrics in incident response because it defines how much time defenders have to detect and contain an intrusion before it expands. In many recent incidents, breakout time is under 30 minutes.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。