AI application security tools agree on just 5 percent of their findings, according to a new report from Contrast Security that paints a grim picture of defenders losing ground against automated attacks. The AppSec Overflow 2026 study draws on telemetry from inside hundreds of thousands of production applications and APIs to quantify what security teams have suspected for months.
Automated Attacks Are Overwhelming Defenses
Adversaries touch the average application once every four minutes. Most of that traffic consists of automated reconnaissance, scanners mapping weaknesses and cataloging services. A smaller slice delivers exploit payloads and attempts to manipulate application behavior. Contrast recorded 42 confirmed, viable exploit attempts per application every month, meaning the vulnerability was not just probed but triggered.
Untrusted deserialization led the field among these confirmed exploits, followed by path traversal and method tampering. SQL injection appeared in the top five attack techniques across every industry vertical the report tracked, from finance to healthcare to manufacturing.
A company with a few dozen employees runs roughly two dozen applications and APIs on average. That number climbs with headcount, and organizations with more than 5,000 employees operate several hundred applications and APIs on average.
Related: Citrix NetScaler Vulnerability Exploited in Wild
Patch Backlogs Stretch Into Years
Applications monitored by Contrast carry an average of 106 vulnerability findings in code written in-house, including 22 rated high or critical severity. Development and security teams close only a handful of these each month. Fixing a highest-severity application vulnerability takes an average of 92 days, and the recorded average remediation rate sits at just 3.4 vulnerabilities per application per month.
Spring4Shell and Log4Shell, both years old, still show up widely in production telemetry.
For twenty years the discipline of AppSec has been organized around a race: find the vulnerability, decide if it matters, and fix it before somebody with bad intent finds it first. AI ended that race, and defenders lost it. We are now seeing vulnerabilities weaponized in hours while the average critical fix takes weeks or months.
The window between discovery and exploitation has compressed dramatically. Zero Day Clock, which aggregates exploit signals from more than 83,000 CVEs, recorded a mean time to exploit of more than two years in 2018. That figure dropped below one year by 2021, and the majority of exploited vulnerabilities in 2025 were weaponized within three weeks.
Related: Google Loses Four Top AI Researchers
Bug bounty programs are feeling the strain. HackerOne paused new submissions to the Internet Bug Bounty program in March 2026, the longest-running crowdsourced vulnerability program in open source. Node.js paused its own bounty program shortly afterward, citing the loss of that funding.
AI Tools Add Complexity, Not Clarity
Contrast set three AI scanners loose on the same codebase and found they agreed on just 5 percent of findings. Running a single scanner three times against identical code reproduced only 17 percent of its own findings. Scanning a 2-million-line codebase with AI tools cost around $315 in API charges. Triaging the resulting findings cost roughly $128,000.
AI is not going to triage its way out of this problem, and we have the data to prove it. These tools disagree with each other; they disagree with themselves from one run to the next, and none of them can tell me how my application behaves when someone is actually attacking it. That is fine when AI is one input among several. It is a problem when it becomes the system of record, because that is what decides what my team works on Monday morning.
