The Packet I Never Sent: Finding Hosts I Could Pop Without Touching Them

Senior Exploit Developer VulnCheck

Vulnerability reports are being submitted faster than maintainers can triage them, and bug bounty programs are struggling under the volume of low-quality, AI-generated submissions. In January 2026, the curl project closed its bug bounty program after fewer than 5% of its 2025 submissions identified a real vulnerability. Google, Apple, and GitHub have all since tightened their submission requirements. Producing a professional-looking report now costs very little and plausible reports take more effort to disprove. This research began with the question: could VulnCheck's data produce the opposite, a small set of findings that merit an organization's attention?

To answer this question, I built a tool that joins exploits written or manually reviewed by VulnCheck, Internet-wide fingerprinting of the software running on each host, and the vulnerability-to-product mapping that connects the two. At no point does it send traffic to a target. The results identify millions of potentially vulnerable hosts, of which only a small fraction belong to organizations that have authorized any form of testing. This is the central limitation the tool exposes. Without testing, every match remains an inference, the same class of unverified report that is overwhelming triage teams today. The exploits exist, as does the data to locate every exposed host they affect. What remains missing is authorization, and only the exposed organizations are in a position to provide it.

‍