JFrog Exposes Fake SQLite CVEs Possibly AI-Generated
Original: SQLite Critical CVEs or LLM Slop?
Why This Matters
AI-generated fake CVEs reaching NVD and CISA ratings highlight critical gaps in vulnerability vetting infrastructure.
JFrog Security Research revealed that 6+ critical CVEs filed against SQLite in a newly created GitHub repo (programmervuln/cveadvisory) appear to be AI-generated fabrications. NVD rated them up to 9.8 Critical, but JFrog's analysis found non-existent functions, invalid PoCs, and contradictory metadata.
JFrog Security Research published findings on July 30, 2026, exposing a batch of SQLite vulnerability advisories from a newly created GitHub repo (programmervuln/cveadvisory) as likely AI-generated ('LLM slop'). Despite NVD flagging several as CRITICAL and CISA's ADP concurring, JFrog's hands-on investigation found the claims baseless.
Key findings across 6 CVEs (CVE-2026-51302 through CVE-2026-51297): cited functions did not exist in the referenced SQLite versions, PoC SQL payloads failed to trigger crashes under AddressSanitizer instrumentation, and none appeared on SQLite's official advisory page. AI-detection tool GPTZero flagged all advisories as AI-generated.
For CVE-2026-51302 (rated 9.8 Critical), the advisory claimed a heap use-after-free in exprComputeOperands()—a function that didn't exist in SQLite 3.41; it was added in mid-2025. Red Hat initially scored it 10.0 Critical before downgrading to 7.6 High.
JFrog's methodology included cloning the official SQLite repository, compiling releases in isolated Docker containers, running PoC payloads with ASan, and auditing NVD/GHSA metadata. JFrog stated it believes the repo's 50+ CVEs are similarly fabricated, with one possible exception.