AI-Generated Fake Vulnerabilities Overwhelm CVE Pipeline as NIST Backlog Balloons
Key Takeaways
- ▸Six fake SQLite CVEs with CVSS scores up to 9.8 were discovered in the NVD, generated by AI systems with fabricated flaws that don't exist in the actual code
- ▸NIST's vulnerability processing backlog has grown to 27,000+ unreviewed CVEs, eliminating the manual verification backstop that previously caught errors
- ▸The CVE pipeline operates on an honor system with no mandatory proof-of-concept requirements—allowing AI-generated garbage to reach developers and enterprises
Summary
A batch of critical and high-severity vulnerability reports in the National Vulnerability Database has been exposed as AI-generated fabrications, exposing fundamental weaknesses in the CVE pipeline. Security firm JFrog discovered that six SQLite CVEs with CVSS scores ranging from 7.5 to 9.8 contained complete fabrications—including non-existent functions, misattributed source lines, and non-reproducible flaws that none of the accompanying proof-of-concept code could validate. At least 50 additional fake CVEs claiming vulnerabilities in RAW image processing and audio decoding libraries also polluted the database.
The crisis has been compounded by NIST's collapsing vulnerability review process. The agency's backlog of unprocessed CVEs has doubled from 17,000 in late 2024 to over 27,000 by the end of 2025, according to a Department of Commerce Inspector General report that attributed the failure to "lack of strategic planning and decisive action." NIST used to provide a reliable verification backstop by manually reviewing and enriching CVE records, but that process has effectively halted. With no mandatory proof-of-concept reproduction requirement and no independent verification checkpoint, plausible-sounding AI-generated vulnerabilities now flow directly into widely-used security databases that developers and enterprises rely on for patch prioritization.
- At least 50 fake CVEs across multiple open-source projects (SQLite, libraw, ESP32-audioI2S) have already been published
- Security researchers are catching these fakes, but with NIST disabled, the burden of verification now falls on individual maintainers and security firms
Editorial Opinion
The CVE pipeline's vulnerability to AI-generated garbage represents a dangerous collision of two institutional failures: NIST's administrative collapse under backlog pressure, and the proliferation of low-cost AI systems capable of generating plausible-sounding technical nonsense. While security researchers are working to clean up the mess, the system's reliance on an honor system and reactive verification is clearly inadequate. Without mandatory proof-of-concept reproduction, significantly increased NIST funding, or stricter CNA (CVE Numbering Authority) vetting, we should expect more fake vulnerabilities to poison the pipeline—wasting organizational security resources on phantom flaws while real vulnerabilities languish unreviewed.
