The AI Exploit Gap: How AI is changing security

AI has made nearly every part of building software faster, and nearly every part of attacking it faster, too. Code ships faster. Dependencies pile up faster. Frontier models now surface critical flaws by the thousands and chain “minor” bugs into working exploits. Attackers use the same tools, so exploitation often starts before a patch even exists.

The one step that hasn’t sped up is the fix. Remediation still moves only as fast as humans can review, and competes with product deadlines and uptime.

We call that mismatch the AI exploit gap: the widening gap between how quickly AI-powered tools can find and chain vulnerabilities and how quickly human teams can respond. To measure it, we commissioned an independent survey of 1,300 director-level and above security leaders across nine markets. The full report, The AI Exploit Gap, is out now. Here’s a preview of what we found.

Three numbers defining the AI exploit gap

• 91% of security leaders say AI-accelerated attacks have changed their job more than anything else in the past decade.

• 77% believe it’s likely an attacker will exploit a software vulnerability at their organization in the next 12 months.

• 54% say their companies knowingly leave high- or critical-severity vulnerabilities unpatched at least sometimes.

Security leaders know the job has changed, and they can see what’s coming. Their organizations’ ability to act hasn’t caught up. The distance between those numbers is what we explored through this report.

"AI has changed the security job faster than anything else in the past decade, and security leaders can see exactly where it's heading. Most we talked to expect an attacker to exploit a vulnerability in their organization this year," said Quincy Castro, Chief Information Security Officer, Chainguard. "But seeing the risk and being able to act on it are two different things. AI can find and chain flaws in minutes, while fixing them still moves at human speed and competes with product deadlines and uptime. That's the AI exploit gap, and it hits security and engineering teams in the areas where many historically struggled prior to AI. Thinking you'll just patch faster is the Red Queen talking. Patching faster isn't viable for most folks and it will consume increasing amounts of resources just to keep pace. The way out is for engineering and security to own the problem together and make prevention the standard, instead of just trying to run twice as fast.”

AI made the job bigger and the clock shorter

AI was supposed to make security teams’ jobs easier. So far, it’s done the opposite. 64% of security leaders say concern about AI-accelerated attacks has added to their own burnout in the past year, and 96% say team burnout affects their ability to defend.

Part of the reason is structural. Engineering and DevOps own the code and the patching, so they’re on the front line. But when a known vulnerability gets exploited, security takes the blame. Being held accountable for risk you don’t control wears people down over time.

“Low severity” no longer means “safe to ignore”

Every unpatched vulnerability is a bet that attackers won’t find it before you fix it, and that bet keeps getting worse. When frontier AI models can chain multiple low-severity bugs into a working exploit, there’s no such thing as a harmless backlog. It’s no surprise that 93% of leaders worry their remediation speed can’t keep up with AI-assisted exploits.

The stakes have also moved past risk tolerance. The EU Cyber Resilience Act now requires reporting actively exploited vulnerabilities within 24 hours, and the EU Product Liability Directive treats software as a product starting December 9. In the US, negligence turns on what a reasonable company should have done. If there’s a record showing you knew about a vulnerability and chose to wait, that’s a hard position to defend to a regulator, an insurer, or a jury.

Big companies have a will problem, not a skill problem

You might expect the largest companies, with the most people and tooling, to be the least exposed. The data says otherwise. Companies with $1B+ in revenue leave high or critical vulnerabilities unpatched at more than twice the rate of companies under $250M (66% vs. 27%).

The reasons are different, too. Smaller companies point to capacity limits. Larger companies point to business trade-offs: they have the resources, but patching keeps losing out to other priorities. The full report breaks down what’s standing in the way at each.

Urgency works like a security control

One of the clearest patterns in the data is that the leaders who feel the clock ticking are the ones actually fixing things. Among leaders who believe AI already outpaces human remediation, 20% leave critical vulnerabilities unpatched at least sometimes. Among those who think that shift is more than a year away, it’s 64%.

That matters because only about half of security leaders (54%) expect AI to routinely outpace human remediation within a year, even as the people building AI warn that the window to defend is narrowing.

Budget isn’t what’s missing

70% of security leaders say securing budget for AI-driven threat defense is easy. Money isn’t the bottleneck. Execution and urgency are. 39% have bought tools they haven’t rolled out, and 93% say their leadership needs to show more urgency about AI-accelerated threats.

Closing the gap

Patching faster after the fact won’t address the gap. The teams that win will get ahead by not shipping the vulnerability in the first place. That means making prevention the standard, building on secure-by-default components, and making remediation a shared responsibility, so engineering teams that own the fix and security teams that respond to the exploit work toward the same outcome.

The full report goes further: why confidence in tools, incident response, and investment doesn’t change how likely leaders think an attack is; how the gap differs by company size, revenue, and seniority; and what security and engineering teams can do now to close it.

Read the full report today.