AI Didn’t Create New Security Problems. It Just Sped Them Up.

Every year around this time, I get some version of the same question from clients: “What’s the new AI threat we need to worry about?” It’s a fair question. Headlines about agentic attacks, autonomous malware, and AI-driven fraud make it seem as if an entirely new category of risk has arrived overnight. But after three decades in this industry, I’d push back on that framing. AI hasn’t handed attackers a new playbook. It’s handed them a faster copy of the old one.

Think about the fundamentals that have always mattered in cybersecurity: who has access to what, how quickly you can spot something unusual, and how fast you can recover when something goes wrong. None of that has changed. What’s changed is the clock speed. Phishing used to take a human hour to draft and send. Now it can be generated, personalized, and deployed in seconds. Malware that once required a skilled developer to adapt can now rewrite itself on the fly. The attack methods are familiar. The pace is what should keep you up at night.

The Speed of Attacks Has Compressed, Not the Tactics

I don’t say this on a hunch. Data from 2025 and early 2026 tell a consistent story: the same categories of attack, executed dramatically faster. Cofense’s threat intelligence team documented a malicious email attack landing every 19 seconds in 2025, more than double the pace measured just a year earlier. That is not a new attack type. It is phishing, the same technique defenders have fought for two decades, now arriving at a volume and cadence that manual triage was never built to handle.

The access side of the equation has compressed just as quickly. Google Cloud’s Mandiant, drawing on more than 500,000 hours of frontline incident response, found in its M-Trends 2026 report that the window between an attacker gaining initial access and handing that access off to a follow-on threat actor has fallen to roughly 22 seconds, down from more than eight hours just a few years ago. Whatever gap used to exist between “we noticed something odd” and “the damage is already done” has effectively disappeared.

The World Economic Forum’s Global Cybersecurity Outlook 2026, produced with Accenture, captures the leadership view of this shift: 94% of executives surveyed now consider AI the single most significant driver of change in cybersecurity for the year ahead. Meanwhile, IBM’s Cost of a Data Breach research found that a striking share of AI-related breaches trace back to gaps in access controls unrelated to novel AI-specific exploits. In other words, the entry point is still the same old door. It’s just getting kicked open faster.

I think many organizations are misdiagnosing the problem, and that misdiagnosis is dangerous. If leadership believes AI security is a brand-new discipline requiring an entirely new toolkit, they end up chasing point solutions and vendor pitches instead of shoring up the basics that were already shaky. I’d rather see companies ask a more useful question: are our identity and access controls strong enough to withstand an attacker who never sleeps, never gets tired, and never makes a typo?

That’s really the heart of it. Credential theft and insider risk have been growing for years, long before agentic AI entered the conversation. Multi-factor authentication, patching, identity monitoring, and least-privilege access aren’t new ideas. They’re the same fundamentals we’ve been preaching since the early days of managed security. What’s different now is that a compromised credential in the hands of an AI-driven tool doesn’t sit quietly for weeks while a human threat actor plans the next step. It moves immediately, at machine speed, probing for the next opening before your team has finished their morning coffee.

Identity Sprawl and Non-Human Identities Expand the Attack Surface

This is also where autonomous AI agents complicate an already stretched identity model. Agents authenticate with API keys, service accounts, and OAuth tokens rather than a username and password, and those non-human identities are multiplying inside enterprises far faster than governance processes were designed to track. Analysis of agentic AI’s effect on identity sprawl has pointed to the same root issue I see in client environments: organizations that already struggled to inventory their service accounts are now minting new machine identities at a pace that makes manual oversight impossible.

Three Practical Steps to Reinforce Operational Fundamentals

So what does that mean practically? First, stop treating “AI security” as a separate line item in your security program. It isn’t a bolt-on. It’s an accelerant for the governance work you should already be doing. If your identity program has gaps, AI will find them faster than a human ever could. If your patch cycle is slow, AI-assisted reconnaissance will exploit that lag before your next change window.

Second, shift your mindset from detection-only to resilience. I say this to clients constantly: assume you will be compromised and build for what happens next. Detection matters, but it can’t be your only line of defense anymore, not when attackers can adapt their approach in real time. The organizations that weather this well are those with tested incident response plans, clean backups, and a clear sense of what “recovery” looks like before they need it. That’s not an AI strategy. That’s just good operational discipline that AI now makes more urgent.

Third, get honest about your own AI-aware threat modeling. That means looking at every system, workflow, and third-party integration and asking a simple question: if an attacker used AI to move faster here, what would break first? For many mid-market companies, the honest answer is uncomfortable. Legacy access policies, sprawling vendor contracts, and manual monitoring processes were tolerable when attacks moved at human speed. They’re a liability now.

There’s a broader industry conversation happening right now about what it actually looks like to govern autonomous systems rather than just the humans who operate them. That conversation matters because agentic AI does not just add a new tool to defend; it changes who or what is taking action within your environment. Treating every AI agent as a distinct identity, with its own lifecycle, entitlements, and behavioral baseline, is quickly becoming table stakes rather than a nice-to-have. The organizations already doing this are best positioned to adopt agentic tools without inheriting an ungoverned attack surface.

2026 has already shown us that AI-driven attacks are real, they’re scaling, and they’re not going away. But I’d rather see leadership teams spend their energy reinforcing the fundamentals than searching for a silver-bullet AI defense tool that promises to solve a problem those fundamentals should have addressed years ago.

The uncomfortable truth is that most breaches still trace back to the same root causes they always have: a missed patch, an over-permissioned account, a gap in monitoring, a plan that was never tested. AI hasn’t rewritten that story. It’s just made the consequences of ignoring it arrive a lot faster.

If there’s one thing I’d ask every business leader to take away from this moment, it’s this: don’t let the word “AI” distract you from the work you already knew you needed to do. Shore up identity. Build for resilience, not just detection. And assume, every day, that the fastest attacker you’ll ever face is one that never has to stop and think.