Security fixes are being bypassed more often in 2026, according to AI penetration-testing startup Terra Security, whose findings The Register reported Monday, Oct. 19. Terra reviewed every published CVE record and GitHub-reviewed advisory looking for broken fixes and found a pattern that had been rare for six years: “A fix would ship, a bypass would be found, and another fix would go out.”
From 2019 through 2025, about 0.2% of GitHub security fixes needed a follow-up advisory reporting a failed patch within 90 days. In 2026 that share jumped to 2.3%. Vulnerabilities whose fixes were later bypassed were also about four times as likely to land on CISA’s Known Exploited Vulnerabilities list as other bugs from the same vendors, Terra said. The Register tied the rise to this year’s boom in AI-driven vulnerability discovery and patching.
One example: Russian government spies exploited a Windows Shell security bypass, CVE-2026-21510, at least a month before Microsoft patched it in February. That patch was incomplete and left a second flaw, CVE-2026-32202, which attackers also found and abused before Microsoft fixed it in April.
Which fixes break
Prototype-pollution and server-side request forgery fixes break most often. Compared with their share of all CVEs, prototype-pollution bugs were about 11 times as common among broken-fix CVEs and SSRF bugs 4.6 times as common, because both are usually patched with a reject-list of inputs such as hostnames, paths and object keys. “If the blocklist doesn’t foresee every encoding of a bad input, the attacker only needs to find one,” the researchers said.
Kernel fixes carry their own risk. Based on kernel CNA records as of Oct. 8, Terra said “3.2 percent of kernel CVEs were introduced by the fix for another CVE, and about 10 percent in busy subsystems like fs/smb and net/sched.”
More bugs, faster exploits
The volume of fixes keeps growing. Microsoft’s October Patch Tuesday on Oct. 13 addressed a record 1,411 flaws, BleepingComputer reported, as the company uses AI to find and fix vulnerabilities in its products. Infosecurity Magazine noted that Microsoft released just over 1,100 updates in all of 2025.
Google Threat Intelligence Group recorded 141 vulnerabilities disclosed and exploited from January through August 2026, more than the 127 exploited in all of 2025, while monthly disclosures doubled from 5,045 in January to 10,740 in August. GTIG said most of the growth in exploitation came from rapid weaponization of already disclosed n-day flaws rather than zero-days.
Microsoft’s 2026 Digital Defense Report, published Oct. 1, said the median time from a vulnerability’s discovery in the wild to weaponization “has fallen to well below 24 hours,” while enterprise remediation for critical external vulnerabilities “can take 30 to 60 days.”
Regulators have been warning about this since spring. India’s CERT-In, in a May blueprint on defending against AI-assisted exploitation, set a target of containing or patching known exploited vulnerabilities on internet-facing systems within 12 hours where feasible. On May 1, Ollie Whitehouse, CTO of the UK’s National Cyber Security Centre, told organizations to prepare for a coming “patch wave” and to identify and minimize their internet-facing attack surfaces as soon as possible.
Handing the fixes to AI isn’t a clean answer either. In an August study, 1Password’s Off-by-1 Labs generated 6,080 patches for six recently disclosed CVEs with two frontier models and found that 53.9% failed to fix the bug, introduced a new vulnerability, or both.
Terra’s advice to vendors is to check that a patch fixes the root cause, covers the whole bug class and doesn’t break something else. For the teams applying it: “If a vendor fixes an exploited bug, assume someone is reading the patch for a way around it.”