AI slop infiltrates the CVE system: "the cited code didn't even exist"
Nothing is sacred for AI slop
By GettyImages
- JFrog found dozens of likely AI-generated fake vulnerabilities, with some receiving official CVE identifiers and Critical ratings.
- Several fabricated advisories targeted SQLite, but researchers said the cited vulnerable code and exploit behavior did not exist.
- The fake entries entered trusted security pipelines, risking false alerts, wasted patching work, and reduced trust in vulnerability data.
- The incident shows AI-generated junk is moving beyond consumer platforms and threatening core cybersecurity systems.
Key Takeaways by nexos.ai, reviewed by Cybernews staff.
Dozens of hallucinated software vulnerabilities have slipped into one of cybersecurity's most trusted tracking systems.
Researchers at JFrog uncovered what they describe as AI-generated "slop" vulnerabilities, many of which were even assigned official Common Vulnerabilities and Exposures (CVE) identifiers.
The researchers traced the issue to a GitHub repository that published 55 CVE advisories, 54 of which they believe were generated by a large language model.
Several advisories targeted SQLite, the widely used embedded database engine, describing critical memory-corruption flaws that could allow arbitrary code execution. There was just one problem: the vulnerable code didn't exist.
One report claimed attackers could trigger a heap use-after-free flaw leading to arbitrary code execution. But when JFrog researchers examined SQLite's source code, they found that the functions referenced in the advisory were nowhere to be found.
"The cited code didn't even exist," the researchers wrote, adding that the accompanying proof-of-concept exploit also failed to produce the crash described in the advisory.
The fabricated advisories still made it into the CVE ecosystem. Several received official CVE IDs and landed in the National Vulnerability Database with Critical severity ratings. CISA's Authorized Data Publisher program even validated their severity before researchers spotted the fake reports.
The incident highlights an uncomfortable reality for the security industry. The CVE program serves as the backbone of vulnerability management, feeding countless security products and patch management systems. Once an inaccurate entry enters that pipeline, it can spread rapidly, lighting up dashboards and triggering automated workflows.
AI slop is showing up everywhere
The SQLite episode is the latest example of AI-generated junk escaping chatbots and creeping into platforms people use daily.
Academic publishing has already been forced to respond. Earlier this year, arXiv announced tougher enforcement against authors submitting papers containing obvious AI-generated filler after repeated incidents of low-quality or fabricated content made it into submissions.
Online marketplaces face a similar problem. Sellers have flooded Amazon, eBay, and Etsy with AI-generated listings for fake flower seeds, using convincing but fabricated images to market plants that do not exist.
Social media platforms are also struggling to keep up. LinkedIn users recently began calling for an easier way to report AI-generated posts as their feeds filled with machine-written engagement bait.
Video platforms have joined the crackdown as well. YouTube recently tightened its policies around mass-produced AI-generated videos, acknowledging growing concerns that low-effort synthetic content was overwhelming original creators.
Until now, most AI slop has largely been an annoyance. A fabricated CVE is different.
The SQLite incident suggests AI slop has moved beyond consumer content platforms. It is now threatening the foundation of cybersecurity's trust infrastructure.
