AI security projects deliver first fixes for vulnerable Java libraries
Chainguard's Athena and IBM and Red Hat's Project Lightwell have reported early Java security fixes, offering DevSecOps teams a potential safety net in the era of AI vulnerability discovery.
As AI makes software flaws easier to find and exploit, two programs using AI to identify and patch vulnerabilities in open source software — Chainguard's Athena and IBM and Red Hat's Project Lightwell — have reported their first results.
Athena, which uses findings generated by AI to identify vulnerabilities, publicly disclosed 14 flaws that had already been fixed in newer software versions but had gone unreported, leaving users of older versions unaware of them. Lightwell used AI to help find and fix vulnerabilities in Java libraries, reporting more than 400 fixes, including developing patches for older versions still running in production. Both sets of results reflect work conducted between July and October.
Experts and DevSecOps practitioners said the projects are a step in the right direction, addressing a persistent software supply chain security problem: vulnerabilities in the open source libraries that enterprises use in their applications, especially those flaws that conventional security scanners might miss. But finding vulnerabilities is only part of the challenge. The findings also need to reach the security tools that enterprises use before they can act on them.
"I'd want to see how long it takes to go from a validated finding to a fix running in production, and how many customers actually deploy it. A fix sitting in a repository nobody pulls from hasn't reduced anyone's risk. I'd also watch how many findings get public identifiers and how many fixes maintainers accept upstream," said Katie Norton, IDC research director.
Katie Norton
Both Chainguard and Red Hat are now CVE Numbering Authorities, organizations authorized to assign identification numbers to publicly disclosed vulnerabilities. Chainguard did not have that designation during most of Athena's initial runs.
Disclosure is a later step in the coordinated vulnerability disclosure process than remediation. Chainguard reports having remediated many more than the 14 disclosed, and is taking the disclosure process slowly so as not to "step on" developers responsible for maintaining the affected open source software, Norton said.
"The industry can now find vulnerabilities far faster than it can coordinate and ship fixes, and enterprises were already behind," Norton said, noting that Chainguard reported more than 21,000 validated findings and 14 public disclosures.
Roughly 40% of enterprises add new findings faster than they can close them each month, according to IDC's DevSecOps and Software Supply Chain Security Survey from July 2026.
Chainguard's Athena is not alone in dealing with the growing gap between finding vulnerabilities and getting them fixed and publicly disclosed. Project Glasswing's recently updated Vulnerability Disclosure ledger showed a similar disparity between the mountain of vulnerabilities AI discovers and the number that humans can triage, remediate and disclose. An Omdia survey showed that AI is expanding the attack surface, with nearly 80% of organizations reporting cybersecurity incidents related to AI threats in the last year.
Both Athena and Lightwell sell patched versions of open source software to organizations that can't upgrade fast enough, Norton said. "They share the same thesis," she said. "AI is finding vulnerabilities faster than the old disclosure model can absorb."
Fixing old Java without breaking apps
Both projects' findings were entirely in Java libraries. Athena focused on vulnerabilities that had already been fixed in newer versions of the libraries but had not been disclosed. Lightwell focused on novel vulnerabilities in both current and older versions of Java libraries used by enterprises.
"Java is where enterprise risk and remediation economics line up. It runs [in] the long-lived, heavily regulated systems in banking, insurance, and government, which are exactly the buyers these efforts courted first," Norton said. "Those systems stay on library versions for years because upgrading is a project."
Companies running older Java libraries are then potentially exposed to vulnerabilities that AI can now uncover — and work backward to exploit — by reading the fixes made in newer versions, Norton said.
"Some of these [findings] are serious, too. One of Chainguard's first disclosures could let an attacker log in as any user," Norton said, referring to CGP-xpq5-jm7p-884r, one of Chainguard's first 14 vulnerabilities, which affects jjwt, a popular Java library.
In Lightwell's case, AI-assisted patches were developed for vulnerabilities in both newer and older Java libraries. By applying those patches to older versions, enterprises can avoid the risks of upgrading legacy software used in mission-critical applications, said Neil Carpenter, advisor at Trent AI.
"Java is where there tends to be the most need for a backport [and] the most resistance to upgrading. So, I think for both projects, it's a sensible place to start," Carpenter said.
Broadcom's Truesource is another initiative tackling the same supply chain security problem. Broadcom maintains Spring, a popular Java development framework.
"Lots of people are pinned to specific versions of Spring," Carpenter said. "There have been a ton of significant vulnerabilities in Spring and parts of that framework over the last five years."
Safety net for DevSecOps teams
While the first fixes are useful, practitioners view the Athena and Lightwell initiatives as ancillary to tried-and-true security best practices.
"My view is that both [initiatives] are useful, but more as a safety net than a must-have for organizations that already follow strong security practices," said Suresh Gangula, a DevOps engineer at Nike.
Suresh Gangula
For a mature engineering team, the preferred fix for a vulnerable library is still upgrading to a supported upstream version, Gangula said.
Teams also use software composition analysis, static and dynamic application security testing, secret scanning and automated security checks in development pipelines to catch vulnerabilities, he said. AI coding assistants and IDE security tools also flag potential problems while code is being written, although those findings still need to be validated by established security controls.
Lightwell and Athena are most useful when those practices fall short, Gangula said, citing two examples: fixes that go unrecognized by scanners because they never got a CVE advisory and cases in which upgrading to a newer version breaks compatibility or compliance.
"In those situations, a vetted backported patch or a rebuilt artifact from a trusted vendor reduces the risk right away and gives the team time to plan the proper upgrade," Gangula said.
A vetted backported patch or a rebuilt artifact from a trusted vendor reduces the risk right away and gives the team time to plan the proper upgrade.
Suresh GangulaDevOps engineer, Nike
A clean dependency scan, however, isn't proof that software is secure, said Norton. "Teams should expect their scanners to start flagging versions that used to come back clean, because those vulnerabilities were there all along." When they don't, teams should remain vigilant. "The operational priority is getting fixes safely into production, even when no scanner has prompted them to act," she said.
Lightwell has the potential to help resolve the tradeoff that dependency management presents for developers and take the pressure off teams, said Kacper Michalik, software engineer at Screen Studio.
"Realizing that one of the packages your project depends on has a serious vulnerability is stressful enough on its own because it's not your team's fault, yet you have to deal with it quickly," Michalik said. "One practice that helps is pinning exact dependency versions wherever possible … so an update can't silently pull in something you didn't test. The tradeoff is that pinning also means you miss automatic security patches -- which is exactly the gap something like Lightwell aims to close."
Ben Lutkevich is an award-winning writer for TechTarget.
Katie Norton
Suresh Gangula
