A surge in reported patch failures has emerged as artificial intelligence tools accelerate vulnerability discovery, according to research by Terra Security. The firm analyzed publicly disclosed Common Vulnerabilities and Exposures records and GitHub security advisories, identifying a sharp increase in cases where patches were later bypassed within 90 days. This pattern marks a departure from a six-year trend of relatively stable patch reliability between 2019 and 2025, when bypass incidents remained below 0.2 percent of all fixes.

In 2026, the rate of broken patches climbed to 2.3 percent, with vulnerabilities patched and later circumvented appearing on the U.S. Cybersecurity and Infrastructure Security Agency’s Known Exploited Vulnerabilities list at four times the frequency of other active bugs. Server-side request forgery and prototype-pollution vulnerabilities were disproportionately affected, accounting for a significant share of bypassed fixes. These classes often rely on reject-list filtering mechanisms that can be circumvented through encoding variations or overlooked edge cases.

Kernel-level patches also exhibited elevated failure rates, with 3.2 percent of Linux kernel fixes introducing new vulnerabilities themselves. Subsystems such as fs/smb and net/sched showed particularly high recurrence of introduced bugs. Researchers noted that patch bypasses frequently stemmed from incomplete remediation of root causes, insufficient coverage of attack vectors, or unintended side effects in fix implementations.

The findings suggest a shift in threat dynamics, where rapid AI-assisted vulnerability identification compresses the window between patch release and exploitation. Terra Security advises vendors to evaluate patches for root-cause resolution, comprehensive class coverage, and potential secondary failures before deployment. Similarly, system administrators are encouraged to treat patched vulnerabilities as potential attack vectors until proven secure, assuming adversaries may analyze patches for bypass techniques.

The trend underscores the need for more rigorous patch validation and proactive threat modeling in software development lifecycles, particularly as AI tools lower the barrier to discovering exploitable flaws.