SecureWorld News

Vulnerabilities Found Outpace Fixes by 18 to 1

Written by Cam Sivesind | Wed | Nov 18, 2026 | 3:21 PM Z

Security teams have never been better at finding vulnerabilities. Fixing them is another story.

Black Duck's new AI Readiness Report for 2027 puts numbers on that gap. The company analyzed 65.5 million application security testing findings across roughly 13,100 applications at 533 organizations, collected between June 2025 and May 2026. The pattern holds month after month: discovery keeps speeding up, remediation doesn't keep pace, and the distance between them keeps growing.

The timing is bad. Black Duck cites Zero Day Clock data showing that in 2018, the mean time from a vulnerability's disclosure to its first exploit was 2.6 years. Today, it's less than a day. Remediation at most organizations still runs on a monthly cycle.

In June 2025, organizations in the dataset found 11 new vulnerabilities for every one they closed. By May 2026, the ratio was 18 to 1, and it peaked at 22 to 1 in December 2025. In April 2026 alone, 10.4 million unresolved findings were added to backlogs.

By the end of the study, only about 1 in 10 findings (10.5%) had been closed.

Black Duck's conclusion is blunt. At these ratios, doubling remediation capacity would still leave the backlog growing. An 18-to-1 problem becomes a 9-to-1 problem, and that's still a losing position. The report says this isn't a staffing problem, because no amount of hiring keeps up with growth measured in millions of findings a month. It calls it an operational scale problem.

The backlog is also getting more dangerous. Critical and high-severity findings made up 16.8% of new findings in May 2026, the highest share of the study year, up from 13.1% in June 2025. Across the full dataset, 9.2 million findings were rated critical or high.

A month used to be fast

Among findings that did get closed, the speed depends heavily on how they were found.

  • Testing method: SAST (proprietary code); median time-to-close: 28 days; share of findings closed: 8.7%

  • Testing method: SCA (open source components); median time-to-close: 39 days; share of findings closed: 15.0%

  • Testing method: DAST (running applications); median time-to-close: 316 days; share of findings closed: 49.2% 

SAST and SCA account for 99.7% of all closures, and they close fastest. But they also close the smallest share of what they find. DAST and manual testing work the other way: they take most of a year per finding, but they close several times more of their findings.

The long tail is worse. Ten percent of DAST findings stay open for 3.8 years or more, and 10% of manual testing findings stay open for three years or more. Even with SAST and SCA, the slowest 10% take about 10 months or longer.

The report's conclusion puts it simply: "A month used to be fast. It isn't anymore."

Severity isn't setting the order

The most uncomfortable finding is about which vulnerabilities get fixed first. Among closed open source (SCA) findings, critical-severity issues took a median of 58 days to close. High-severity issues took 22 days. In SAST, high-severity findings closed more slowly than medium- or low-severity ones.

Black Duck acknowledges that part of this reflects harder fixes. A critical flaw in an open source component may require a major version upgrade, while a low-severity fix is often a simple patch. But the report's reading is that when findings arrive by the millions, teams fix what's easiest, and nothing deliberate is setting the order.

The low-volume methods tell a different story. DAST and manual testing are slower overall, but critical findings close first. The prioritization problem is concentrated exactly where the volume is.

Company size shows a split, as well. Organizations with more than 10,000 employees closed critical and high-severity findings in a median of 12 days, compared with 19 days for all their findings. Some smaller groups showed the reverse: companies with 2,501–5,000 employees took 28 days on their most serious findings and 21 days on everything else. The slowest segment overall was midsize organizations with 1,501–2,500 employees, at a median of 56 days across all findings.

Severity scores alone don't settle the question, either. In the vulnerability advisories Black Duck's researchers published in 2025, known exploits existed for 45% of high-severity vulnerabilities, 39% of medium-severity ones, and 31% of low-severity ones. A medium-severity flaw with a working exploit in reachable production code can be more urgent than a theoretical critical one.

Read the medians carefully

Jason Soroko, Senior Fellow at Sectigo, cautions against reading too much into the closure-speed figures. "Black Duck researchers found that 85% of open source findings remained open at the study's cutoff," he said. "The 39-day median describes only findings that were closed. Teams need to track the age and exposure of the unresolved backlog alongside closure speed. The longer closure time for critical flaws should prompt a review of upgrade barriers and ownership. That timing alone cannot prove teams chose easy fixes over greater risks."

It's a fair point, and it applies to more than one chart. Time-to-close statistics only count what got closed, so the hardest and riskiest items still sitting in the backlog aren't in the medians.

The report's own methodology adds other limits. The dataset is made up of Black Duck customers who already pay for commercial AppSec tooling and may be more mature than the broader market. Organizations that remediate well may be more likely to stay in the dataset. And Black Duck says its findings on scan frequency, company size, and remediation speed are correlational, not causal.

Coverage gaps leave room for chained attacks

Coverage data adds another layer. Two-thirds of applications (66%) were tested with only one methodology. Only 0.4% were tested with three, and none were tested with all four. Black Duck notes that many enterprises run tools from several vendors, so this doesn't measure total coverage. Still, each method sees a different slice of risk.

Manual testing was the least used method by far, applied to just 475 applications. Yet 53.1% of what it found was critical or high severity, compared with 9.5% for SAST. Those are mostly business logic and authorization flaws, the kind of weakness the report says frontier AI models can now chain with other bugs into multistep exploits.

Open source is a big part of the risk. SCA turned up 18.9 million vulnerabilities in open source and third-party components, which was 52% of all critical and high-severity findings in the dataset. Much of the most serious exposure sits in code the organizations didn't write.

That makes another 2026 development more important. On April 15, U.S. NIST moved the National Vulnerability Database (NVD) to risk-based enrichment. Priority now goes to CVEs in CISA's KEV catalog, software used by the federal government, and critical software as defined under Executive Order 14028. Other CVEs are still listed, but they aren't scheduled for immediate analysis. Black Duck's advice is direct: don't rely on the NVD alone.

The Mythos factor

Black Duck named its framework for this report after Anthropic's Claude Mythos Preview. The report cites Anthropic lab tests in which Mythos Preview produced 181 working Firefox exploits under the same conditions where its predecessor produced two. It also points to Sysdig's analysis, published in February 2026, of an AI-assisted AWS intrusion that went from stolen credentials to administrative access in about eight minutes.

Black Duck expects comparable capabilities to be widely available in AI models within six to 12 months. Its Mythos Readiness Framework has four pillars:

  • Visibility: Do you know what components and vulnerabilities are in your applications?

  • Coverage: Are you testing every application with the methods it needs?

  • Prioritization: Are you fixing what matters most first?

  • Remediation velocity: Are you closing vulnerabilities faster than they accumulate, and faster than attackers can exploit them?

The prescription is to pair deterministic testing (SAST, SCA, DAST) with AI-assisted triage that weighs exploitability, reachability, and business context. Agentic remediation then generates fixes and pull requests, with validation in the pipeline.

Agentic remediation brings its own risk

Handing remediation to AI agents solves one problem and creates another. Aviv Nahum, Co-founder and CEO at Above Security, puts it this way: "Agentic remediation fixes the backlog by hiring a new insider. An agent that can write and merge code to production has commit rights, and it acts at machine speed. If you don't watch what it does between the ticket and the merge, you haven't removed the risk, you've moved it into the pipeline."

Nahum suggests a test before scaling up. For every agent, teams should be able to say "what it has access to, what it's actually doing with that access, and whether that just changed."

His co-founder, Amir Boldo, Chief Product and Technology Officer at Above Security, frames agents in terms any engineering manager would recognize. "Think of each agent as a new engineer with production access," Boldo said. "You'd never give a new hire commit rights without knowing who they report to and reviewing what they ship. Agents need the same: a named owner for every agent, and their changes reviewed like anyone else's."

Black Duck's report agrees with that point. It recommends that automated fixes be rescanned, tested, and submitted as pull requests for a developer to approve, and that people stay in the loop for risk validation, complex fixes, and business tradeoffs.

The regulatory clock is ticking too

The report connects the remediation gap to compliance pressure that is already in force.

  • PCI DSS 4.0.1 requires critical security patches within one month of release. In the dataset, nearly half of the critical open source vulnerabilities that were fixed took longer than 30 days, and most were never fixed during the study period.

  • EU Cyber Resilience Act reporting obligations, which took effect in September 2026, require an early warning within 24 hours for actively exploited vulnerabilities.

  • DORA requires EU financial entities to show how they identify and address vulnerabilities.

  • SEC rules require public companies to disclose material cybersecurity incidents within four business days of determining they're material.

The report also raises the governance question boards will ask after a breach traced to a known flaw: how long was this known, and why wasn't it fixed?

What to do now

Taken together, the report's recommendations and the experts' advice give security leaders a practical starting list.

  • Know what you run. Keep a current SBOM for every application and update it continuously, not just at release. Soroko's starting point is a current inventory of deployed components and which vulnerable code attackers can actually reach.

  • Check your own order of operations. Are your critical findings actually closing before your low-severity ones? If not, find out why. It may be upgrade barriers, unclear ownership, or no deliberate prioritization at all.

  • Prioritize by exploitability, not CVSS alone. Weigh known exploitation, reachability, and business impact. Give each fix an owner and a deadline. Set timeframes for lower-severity findings too, because attackers can chain them.

  • Mitigate when you can't patch. Soroko advises restricting access or disabling the affected feature when an upgrade needs time.

  • Track the right three numbers together: median time-to-close, percentage of findings closed, and discovery-to-closure ratio. A fast median means little if only a small fraction of findings ever gets fixed. Soroko adds a fourth: how long exploitable systems stay exposed.

  • Close only what's verified. Mark a finding closed after confirming the change reached production and removed the exposure.

  • Govern your agents like engineers. Give each one a named owner, scoped access, monitored behavior, and reviewed changes.

The organizations doing best in Black Duck's data aren't facing less risk. They've built pipelines that turn findings into fixes without waiting on manual handoffs. As AI shrinks the time between disclosure and exploitation, that operating model is becoming the minimum for keeping up.