Critical CVE issued for hallucinated SQLite vulnerability

A newly created GitHub repository, programmervuln/cveadvisory-, published a batch of SQLite vulnerability advisories as part of a larger set of more than 50 CVEs that JFrog security researcher Afek Berger and colleagues assessed as largely LLM-generated, with only one exception. NVD marked the SQLite entries critical, and CISA’s ADP agreed. When JFrog dug into the claims, they did not hold up. The findings were reported around 30 Jul 2026 and later surfaced on Hacker News.

On verification, the cited code either did not exist in the versions named or pointed at unrelated logic. Proof-of-concept payloads failed to crash or otherwise trigger the claimed issues. None of the advisories appear on SQLite’s official advisory page, which remains the authoritative place to track real SQLite vulnerabilities. Gptzero flagged the advisories as AI-generated, and concatenating them into a single file produced AI-generated content warnings.

For engineers and security teams, automated severity from NVD and CISA ADP is not a substitute for package-level proof. A critical label can force ticket storms, emergency patches, and false confidence that a real remote crash path exists. Builders who wire scanners and ticket systems to CVE feeds without checking vendor advisories and runnable PoCs will burn time on noise while real issues wait.

The episode sits in a market where CVE assignment, NVD enrichment, and ADP agreement are treated as trust signals even when the originating “advisory” is a fresh GitHub dump. SQLite’s own advisory process still acts as a filter that bulk LLM-style repos do not pass. That gap between pipeline trust and primary-source verification is what let fabricated SQLite criticals look real long enough to matter operationally.

Treat unsolicited CVE batches against well-audited libraries as untrusted until the official vendor page lists them and a PoC reproduces on the claimed version. Watch how NVD and ADP handle challenge or withdrawal after JFrog-style rebuttals, and whether scanners add stronger checks for vendor confirmation and non-reproducing PoCs before elevating “critical” from new advisory repos.