AI Speeds Up Flaw Discovery, Outpacing Fixes

A Black Duck report finds organizations are discovering far more software vulnerabilities than they can close, with AI exploit tools raising the stakes.

Updated on Nov. 20, 2026 in Cybersecurity

AI Speeds Up Flaw Discovery, Outpacing Fixes

Black Duck's 2027 AI Readiness Report, analyzing 65.5 million software vulnerabilities across 533 organizations between June 2025 and May 2026, found companies discovering 18 new flaws for every one they fixed by May 2026. Only 10.5% of flaws had been fixed by the end of the study.

Why it matters

The gap between discovery and remediation leaves growing numbers of unpatched weaknesses in software that organizations depend on. Black Duck expects AI-driven exploit capabilities to become widely available within six to 12 months, which could shorten the window attackers need to exploit these flaws.

Open source and third-party components accounted for 52% of critical and high-severity vulnerabilities, yet only 15% of open source flaws had been fixed. Lab tests showed Anthropic's Claude Mythos Preview producing 181 working exploits against the Firefox browser.

The players

Black Duck

Black Duck is a software security company that provides application security testing and open source risk management tools to enterprises.

Anthropic

Anthropic is an artificial intelligence company that develops the Claude family of AI models.

The details

Black Duck analyzed aggregated, anonymized data from its commercial application security customers. Companies with more than 10,000 employees closed critical and high-severity flaws in a median of 12 days, while firms with 1,501 to 2,500 employees needed a median of 56 days to fix all flaws.

Timeline

  1. June 2025: vulnerability data collection begins.

  2. May 2026: data collection ends, with the 18-to-1 discovery-to-fix ratio measured.

  3. Black Duck projects comparable AI exploit capabilities becoming widely available within six to 12 months.

Seasonal Patterns

The report reflects a documented shift in which the average time from vulnerability disclosure to exploitation dropped from 2.6 years in 2018 to less than a day today. This compression leaves organizations with far less time to patch than the remediation pace measured in the study allows.

Users of open source libraries and third-party software face a backlog of unpatched flaws, since only 15% of open source vulnerabilities had been fixed by the end of the study. Organizations should prioritize updating their dependencies as remediation capacity remains well behind discovery.

The takeaway

Patching cannot keep pace with discovery, so organizations should focus first on the flaws in open source and third-party components, which make up the majority of serious vulnerabilities. Reducing exposure now matters more as AI tools shorten the time between disclosure and exploitation.

Further reading

For more on how attackers and defenders are responding, see Cybersecurity.

Source note: This article includes information reported by TechRound.