Artificial intelligence has already changed how teams build software and automate their work. Now it is changing how they find the security holes in that software. In its Q2 2026 report, Gartner ranks AI-driven vulnerability discovery as the number one emerging risk for organizations, ahead of agentic AI, data integrity, and a lack of readiness for AI itself.
The 20-second version
- For the first time, Gartner puts AI-based vulnerability discovery at the top of its emerging-risk list.
- The survey gathers answers from 316 executives and risk officers.
- Agentic AI climbs to third place.
- The risk shows up when detection runs ahead of validation and patching.
- For IT departments, it points toward more continuous, automated vulnerability management.
The report matters to system, security, and development leaders because of what it says about scale. AI tools can read through large amounts of code, spot vulnerable patterns, and take over parts of a job that used to cost specialists hours.
That helps defensive teams too. The trouble starts when finding bugs gets faster than an organization can check them, rank them, and ship the fixes.
The bottleneck is shifting from finding bugs to fixing them
Gartner’s report draws on a survey run during April and May 2026 with 316 senior risk officers and executives across different sectors and regions.
The results lean heavily toward technology:
| Position | Emerging Risk | Previous Position |
|---|---|---|
| 1 | AI-based vulnerability discovery | New entry |
| 2 | Geopolitical energy supply crises | New entry |
| 3 | Agentic AI | 4 |
| 4 | Data integrity | 1 |
| 5 | Lack of preparedness for AI | 5 |
Topping the list doesn’t mean any model can find a critical vulnerability and weaponize it on the spot.
The work still happens in stages: locate behavior that looks vulnerable, prove there is a real problem, pin down which versions are affected, study whether it can be exploited, and build a fix.
What is new is that parts of this can now be automated at a scale manual analysis alone can’t reach.
Google has documented this. It has used AI systems to find vulnerabilities in software projects through efforts like Big Sleep and OSS-Fuzz, and it is building CodeMender, which puts AI agents to work repairing them.
There is an apparent paradox here: better tools for finding flaws can, at least at first, push up the count of known vulnerabilities.
From a system administrator’s point of view, that isn’t bad news. Catching and fixing a bug before someone exploits it beats leaving it buried for years in a library that thousands of applications rely on.
The real question is what happens when you can discover vulnerabilities faster than you can fix them.
From patch cycles to continuous security
For years, plenty of companies have managed vulnerabilities on a predictable schedule: regular scans, risk ratings, testing, maintenance windows, and rollouts.
AI could break that rhythm.
Once agents start reviewing repositories, dependencies, firmware, and applications around the clock, the pile of potential findings can grow fast. Security teams then have to work out which ones are actual risks.
That puts prioritization front and center.
A critical vulnerability in a library sitting on hundreds of servers might not be exploitable in a given configuration. A lower-rated bug in a service exposed straight to the internet might need fixing today.
Severity scores still help, but only alongside data on exposure, configuration, known exploits, and how important the asset is.
Gartner’s advice is to speed up response, sharpen remediation, and rethink how risk gets assessed when discovery never really stops.
For IT departments, that likely means tying together asset management, software inventory, vulnerability analysis, development, and patch automation.
It also raises the stakes on knowing exactly what software runs in your infrastructure. Spotting a vulnerability helps little if it takes days to figure out which servers use the affected library.
AI helps the defenders too
The same technology worrying risk officers can also shore up security.
AI agents can review code, analyze dependencies, flag risky configurations, generate tests, and suggest fixes. That opens the door to finding and fixing at the same time.
The technical goal is a cycle that runs with less and less manual work:
detect → validate → prioritize → fix → test → deploy
People still matter. On critical infrastructure or business applications, someone has to confirm that a patch doesn’t break something else.
Updating a library in a modern web application is nothing like changing industrial software, legacy systems, or platforms that can’t stop running.
Which brings up an old headache for sysadmins: legacy software.
Many organizations still run applications built years ago on outdated libraries, unsupported versions, or components that are hard to update. Automatically turning up more vulnerabilities in those environments doesn’t make them any easier to resolve.
When you can’t patch right away, other measures carry more weight: network segmentation, service isolation, least privilege, traffic controls, and a smaller exposed surface.
Agentic AI sits high on the risk list, too
The report flags another shift for the tech sector: agentic AI moves from fourth place to third among emerging risks.
Here the worry changes.
Agents can use tools, open applications, query databases, and take actions. The more autonomy they get, the more permissions and control mechanisms matter.
For system teams, it revives a familiar question: what can a given identity actually do inside the infrastructure.
An agent with too many permissions becomes a liability, whether through an attack, a misconfiguration, or a bad instruction.
The old principles still apply: least privilege, separation of duties, activity logging, temporary credentials, and monitoring for sensitive operations.
What is different now is that identities won’t only stand for people, services, and apps. Companies will have to manage agents that can run whole sequences of operations on their own.
Gartner puts data integrity in fourth place and a lack of AI readiness in fifth. Three of the top five emerging risks tie directly back to this technology.
For IT departments, the takeaway is less futuristic than it sounds. AI is speeding up processes that already exist: finding bugs, analyzing software, making decisions, and taking action. The infrastructure has to answer by speeding up its own controls to match.
Frequently Asked Questions
Can AI find software vulnerabilities on its own?
Systems already exist that help locate vulnerabilities and risky behavior in code. But finding a possible flaw doesn’t mean it can be exploited, and someone still has to validate it.
Why does Gartner treat this as a risk?
Because the automated ability to discover vulnerabilities can outrun the processes companies use to analyze and fix them.
Can AI also fix vulnerabilities?
Yes. One active area uses agents to analyze code, propose patches, and check the fixes. Google is among the companies exploring it.
What does this mean for system administrators and IT teams?
It raises the need for accurate inventories, a clear picture of software dependencies, prioritization based on real exposure, and less time spent testing and deploying updates.
