FedRAMP 2026 vulnerability deadlines, worked through an example scan
For years, FedRAMP vulnerability deadlines were a lookup on scanner severity: High in 30 days, Moderate in 90, Low in 180. Monthly continuous monitoring was largely the work of tracking those clocks in a POA&M.
The Consolidated Rules for 2026 end that model. The deadline now depends on what the vulnerability could actually do in your environment. This post walks through the new logic, what it does to a real set of findings, and how to turn a scan export into a first pass at the answers. For the wider set of changes, see FedRAMP's 2026 rules: what changes for continuous monitoring.
Three questions replace one
Under the Vulnerability Evaluation and Reporting rules, a provider must evaluate every detected vulnerability for three things:
| Question | Rule | Short name |
|---|---|---|
| Is it likely exploitable? | VER-EVA-ELX | LEV |
| Is it internet-reachable? | VER-EVA-EIR | IRV |
| What is the potential impact on agency customers, N1 to N5? | VER-EVA-EPA | PAIN |
Two details are easy to miss.
Internet-reachable is not internet-accessible. A service with no route to the internet still counts if a payload from the internet can reach it indirectly. FedRAMP's own examples are SQL injection through an application tier, and Log4Shell triggered deep in a stack by a logged string.
Impact is about customers, not CVSS. N1 is a minimal effect on agencies using the service. N5 is a debilitating effect on more than one agency. A critical CVSS score on a host that holds nothing does not make an N5.
The deadline comes from all three
The timeframe to reduce a finding's impact rating is set by its N-rating, whether it is likely exploitable, whether it is internet-reachable, and the offering's Certification Class. This is rule VDR-TFR-PVR. For Class C, in days:
| Impact | Likely exploitable, internet-reachable | Likely exploitable, not reachable | Not likely exploitable |
|---|---|---|---|
| N5 | 2 | 4 | 16 |
| N4 | 4 | 8 | 64 |
| N3 | 16 | 32 | 128 |
| N2 | 48 | 128 | 192 |
Class B is more lenient (4 days at the top left) and Class D stricter (12 hours). N1 has no fixed timeframe.
Three more rules sit on top of the table:
- 192 days is the ceiling. Anything not fully mitigated or remediated within 192 days of evaluation must be reported as an accepted vulnerability (VER-TFR-MAV). This is the replacement for the long-lived POA&M item.
- Some findings are incidents. A likely exploitable, internet-reachable finding rated above N3 should be treated as a FedRAMP Reportable Incident for Class C and D until it is mitigated to N3 or below (VER-TFR-IRI).
- KEV has its own clock. Known Exploited Vulnerabilities should be remediated by CISA's due date even if they have been fully mitigated (VDR-TFR-KEV).
What the same scan looks like under each model
Here are five findings from an example scan, evaluated under both models.
| Finding | Old deadline | 2026 evaluation | 2026 deadline (Class C) |
|---|---|---|---|
| Log4Shell on an internet-facing host holding multi-agency data | 30 days | Likely exploitable, reachable, N5 | 2 days, and a reportable incident |
| Log4Shell on an internal host | 30 days | Likely exploitable, not reachable, N4 | 8 days |
| OpenSSH Terrapin (CVSS 5.9) on that same internet-facing host | 90 days | Likely exploitable, reachable, N3 | 16 days |
| OpenSSH Terrapin on an internal low-impact host | 90 days | Likely exploitable, not reachable, N2 | 128 days |
| Self-signed certificate, internal, open since January | 90 days, long overdue | Not likely exploitable, N2 | 192 days, now an accepted vulnerability |
The old model gave both Log4Shell findings the same 30 days and both Terrapin findings the same 90. The new one separates them by a factor of four and eight. A "Medium" on the wrong host now has a 16-day clock, which no severity-based process would have produced.
Proposing each answer from a scan
A scan export contains none of the three answers directly. A first pass can still be automated, as long as every value records where it came from.
Likely exploitable. A reasonable starting rule is yes if the CVE is in CISA's KEV catalog, if its EPSS score is above a threshold you set, or if the scanner reports an available exploit. FedRAMP explicitly declines to prescribe a framework here, so whatever rule you use is yours to justify. If exploit intelligence cannot be fetched, assume the finding is likely exploitable. Reporting "not likely" with no evidence is the outcome the rule warns about.
Internet-reachable and asset impact. No scanner knows these. They have to come from asset context the provider maintains, even something as small as a file:
host,internet_reachable,impact
203.0.113.0/24,yes,N4
10.20.1.15,yes,N5
10.20.2.0/24,no,N4
10.20.3.*,no,N2
Hosts missing from that context need a default the analyst chooses, and every value derived that way should be marked as assumed.
Impact rating. One workable approach is the lower of two limits: the asset's impact rating, and a cap from scanner severity (Critical N5, High N4, Medium N3, Low N2). The asset limits how bad it can be; the severity limits how much of that this particular finding could deliver.
The deadline is then a table lookup:
# {class: {PAIN: (LEV + IRV, LEV + not IRV, not LEV)}}
PVR_DAYS = {
"B": {5: (4, 8, 32), 4: (8, 32, 64), 3: (32, 64, 192), 2: (96, 160, 192)},
"C": {5: (2, 4, 16), 4: (4, 8, 64), 3: (16, 32, 128), 2: (48, 128, 192)},
"D": {5: (0.5, 1, 8), 4: (2, 8, 32), 3: (8, 16, 64), 2: (24, 96, 192)},
}
def due_days(cert_class, pain, lev, irv):
row = PVR_DAYS[cert_class].get(pain)
if row is None: # N1: no fixed timeframe
return None
if not lev:
return row[2]
return row[0] if irv else row[1]
What automation cannot do here
The rules require the provider to evaluate each vulnerability in the context of the cloud service. A heuristic over a CSV does not meet that bar on its own:
- Everything automation produces should be treated as a proposal for an analyst to confirm, with the basis shown for each value.
- Scan data records when a finding was first discovered. The rules count from completed evaluation, which a scan does not record. Counting from discovery errs on the strict side.
- The severity cap is a simplification. A medium-severity flaw that exposes credentials for a high-impact system deserves a higher rating than the cap gives it. That is exactly the judgment the analyst is there to make.
- Any export built for the reporting rule needs to be validated against FedRAMP's published schema.
What automation does well is the part that used to consume the week: merging scanner formats, applying the lookup consistently, and showing which findings moved. That leaves the analyst's time for the evaluation itself, which is where the 2026 rules want it spent.
Sources
- Vulnerability Evaluation and Reporting
- Vulnerability Detection and Response
- CISA Known Exploited Vulnerabilities catalog
Rules are as published on fedramp.gov on October 1, 2026, and are still being revised.