Found ≠ Fixed: The Gap Between Vulnerability Discovery and Actionability
Independent researchers are incentivized to identify vulnerabilities. Vendors and bug-bounty triage processes, however, are incentivized differently, and it is in this divergence that many critical findings fail to progress toward remediation.
This presentation examines that divergence, drawing on more than fifteen independent vendor research campaigns and over three hundred findings across network appliances, endpoint detection and response (EDR) and antivirus (AV) agents, and enterprise software. The analysis is grounded in documented case evidence rather than anecdote, and considers four representative cases:
Silent patching followed by rejection, documented on the record. A consumer router vendor released a firmware update addressing a pre-authentication credential-disclosure vulnerability, with release notes stating only that "various security issues" had been fixed;no CVE was assigned, no advisory was published, and no researcher credit was given. When the same vulnerability class was subsequently reported on a related product, the triaging analyst closed the report, stating: "Since this issue was fixed before your report was submitted, it is considered a known issue... this does not qualify for a bounty." This constitutes a documented instance in which silent patching was used as grounds to decline coordinated disclosure.
The evidentiary burden imposed by attack-scenario requirements. This case examines why mature vulnerability disclosure programs increasingly decline technically valid root-cause findings that lack a demonstrated privilege-boundary crossing, on the reasoning that "root-to-root" access does not constitute privilege escalation, and considers the resulting effect on which vulnerabilities are ultimately remediated versus merely discovered.
Reclassification of demonstrated abuse as intended functionality. A widely deployed AI-powered customer-messaging platform permitted an authenticated integration to register a webhook that intercepted the entirety of a customer's support conversations, including authentication tokens embedded in message content, and to silently alter message-routing configuration without administrative notification. The vendor closed the report as "by design," on the basis that webhook registration is an intended platform capability. This case illustrates a mechanism by which a technically substantiated, live-confirmed disclosure vector is rendered non-actionable not through denial or an evidentiary bar, but through characterization of the vulnerable behavior itself as intended.
Self-examination of automated triage tooling. This case presents an account of identifying a failure in the presenter's own AI-assisted triage pipeline, in which a batch of false positives was automatically promoted to "vetted" status, temporarily inflating publicly reported finding counts prior to correction.
This presentation does not argue that coordinated disclosure is fundamentally broken. Rather, it offers a practitioner's account of where the vulnerability discovery-to-remediation pipeline fails in practice, together with a proposed set of criteria distinguishing findings that are merely technically correct from those that are genuinely actionable.