Back to blog

26,502 Findings, 38 Fixes: Why Verified Closure Is the Product

3 min read
26,502 Findings, 38 Fixes: Why Verified Closure Is the Product

Sources checked: . FailSafe Research.

Finding more vulnerabilities will not create a durable security advantage. Microsoft's FORGE Lab reported an AI-assisted Linux kernel campaign that produced 26,502 raw findings in about two weeks, 11,769 screened candidates, 621 proof-of-concept-backed cases, 59 upstream reports, 33 assigned CVEs, eight working root privilege-escalation exploits and 38 merged fixes. Turning scan output into reproducible reports and patches took more than four person-months and continued for about five months.

The exact conversion rates are specific to kernel research. The imbalance generalizes. Discovery tools can now generate credible candidates faster than humans can deduplicate, assign severity, coordinate disclosure, patch and regression test. A product that only increases the first number adds work downstream.

What the numbers show

Roughly 2.3% of raw findings became proof-of-concept-backed cases. Fewer than 0.2% became merged fixes. Those ratios are not a criticism of the researchers, who did the difficult verification work. They show where the time went. Screening, reproduction and disclosure consumed the effort, not initial discovery.

Operating metrics that describe closure rather than volume
MetricDefinitionWhy it matters
Findings to confirmed exploitShare of reported findings with a reproduced impact.Separates candidates from usable results.
Root-cause consolidationFindings grouped by shared cause.Prevents duplicate tickets and repeat fixes.
Human review minutesReviewer time per confirmed issue.Shows the real cost of verification.
Time to accepted remediationConfirmed issue to merged fix.Measures engineering throughput.
Time to verified closureMerged fix to a retested denial.Measures whether risk actually fell.
RecurrenceClosed issues that return after change.Shows whether closure held.

Buyers should price the back half

A pentest or agentic platform that reports a large finding count but no closure data shifts the hard work to the customer. Ask how many findings were rejected after review, how duplicates were consolidated, who owns remediation and whether the original case was retested on the deployed fix. Those questions usually separate a scanner with an agent wrapper from a program that closes risk.

They also change how a security team reports upward. "We found 400 issues" is a volume statement. "We confirmed 37, closed 31 and verified 29 on the deployed build" is a risk statement. A board can act on the second.

Where FailSafe sits

FailSafe SWARM connects authorized attack-path testing with reproduced findings, remediation retesting and evidence handoff to engineering systems. For agreed connected engagements, the workflow can reach remediation pull requests within 48 hours of full access and integration readiness, with customer review before merge. The retest and closure record describes the artifact, and the board metrics guide covers reporting.

This is a product position, not a measured comparison with the FORGE Lab campaign. Microsoft's research concerns Linux kernel vulnerabilities. FailSafe's work concerns applications, APIs, cloud, identity and AI systems in an authorized scope.

Questions & answers

Questions about this topic

Direct answers to common questions about this page.

No. The figures come from one AI-assisted Linux kernel campaign. The discovery and remediation imbalance is the transferable lesson, not the exact ratios.

Time to verified closure: the interval between a merged fix and a retest that shows the original case is denied while legitimate operations still work.

Not without triage and closure. Unverified volume shifts screening and reproduction work to the customer and obscures real risk.

No. The 48-hour figure describes the connected workflow after full access and integration readiness. Customer review and merge remain with the customer.

Put this guidance to work

Review how authorized testing, reproduced findings, remediation handoff and retesting connect.

See the SWARM closure workflow