Not every CVE record describes a real, exploitable vulnerability, and teams that treat every published record and every critical score as fact burn remediation capacity on findings that do not matter while real exposure waits. This guide covers the records the CVE Programme itself flags as contested: what the "disputed" tag means, how a supplier or researcher contests a record under the CVE Record Dispute Policy, how scanners keep alerting on disputed and rejected IDs, and how to handle a severity score you believe is wrong. It also gives consumers a repeatable way to close these findings with evidence an auditor will accept. Programme structure, CNA scopes and the NVD enrichment backlog are covered in the CVE Programme and NVD Backlog Guide; this page goes deeper on contested records only. It is written for vulnerability management leads, PSIRT staff and open source maintainers. Rule references are to CVE Program CNA Rules version 4.2.1 and Dispute Policy version 2.0.0, as checked in November 2026.
Key takeaways
- Disputed is not rejected. A disputed record stays published with an explanation; the CVE Programme makes no judgement on which party is right, and the tag can stay indefinitely.
- Rejected means do not use. A REJECTED record remains on the CVE List only so that people know the ID is invalid; close any finding that cites it and ask the tool vendor why it still fires.
- Disputes follow a written, timed process. Under Dispute Policy 2.0.0 the adjudicating CNA acknowledges within three business days and decides within five more, with escalation to a Root or TL-Root.
- Severity disagreements are not disputes. The base score is a starting point; CVSS v4.0 puts Threat and Environmental scoring on the consumer, which is where most inflated scores should be corrected.
- Close contested findings with evidence. Supplier statements, VEX documents, code paths and configuration proof belong in the ticket, with a review date, not a silent suppression.
The three record states and the disputed tag
The CVE Program glossary defines three states for a CVE Record. Disputed is not a fourth state: it is a CNA tag applied to a Published record. Understanding the difference stops the most common triage error, which is treating a disputed record as withdrawn or a rejected one as merely contested.
| Condition | What it means | Where it shows in CVE JSON 5 | Consumer action |
|---|---|---|---|
| RESERVED | A CNA has reserved the ID; details are not public. An ID used publicly while still reserved is "Reserved but Public". | cveMetadata.state is RESERVED; no description | Track the supplier advisory; nothing to score yet |
| PUBLISHED | The CNA has populated and published the record | state PUBLISHED; cna container with descriptions, affected, metrics | Normal triage |
| PUBLISHED with "disputed" tag | One party disagrees with another party's assertion that the issue is a vulnerability; the explanation must appear in the record | cna.tags contains "disputed"; description usually carries a NOTE stating the disagreement | Read both positions, assess against your own configuration, record the decision |
| REJECTED | The ID and record should no longer be used | state REJECTED; rejectedReasons; optional replacedBy listing the CVE IDs it was rejected in favour of | Close the finding; if replacedBy is set, re-triage against the replacement ID |
The schema's CNA tag list has three values: disputed, unsupported-when-assigned (all affected products were end of life when the ID was requested) and exclusively-hosted-service (the flaw exists only in a hosted service, so there is nothing for a customer to patch). Rule 4.5.3.11 requires CVE Program members to show the disputed status and its explanation when they display record data, so a feed or scanner that drops the tag is not presenting the record as the programme intends.
Why records get rejected
Rule 4.5.3.5 obliges CNAs to reject IDs that were reserved but never used. Rule 4.5.3.6 lets them reject published records, for example when no vulnerability exists, the wrong ID was used, the same flaw was assigned twice, or several IDs were merged into one. A rejection must carry an explanation (4.5.3.7). When IDs are merged, the redundant ones must be rejected, and the CNA weighs publication date and how widely each ID is already used when choosing which survives (4.5.3.8). The NVD status model, including its Rejected status, is explained in How to Read a CVE Advisory Guide.
Why bogus and contested records exist
A CVE ID is an identifier, not a verdict. Several structural features of the programme produce records that a supplier would not accept.
- Right of first refusal has limits: under rule 4.2.1.2, for a publicly disclosed issue, if the CNA with the most appropriate scope declines or does not answer within 72 hours, a Root makes the vulnerability determination and can direct a CNA of Last Resort (CNA-LR) to assign. Projects that are not CNAs can learn of a record after it is published.
- Different vantage points: version 4.2.1 of the rules notes that a supplier CNA's determination may differ from another CNA's, because the latter lacks internal product and engineering information. Rule 4.1.1 says CNAs should agree when a supplier determines a vulnerability exists; there is no matching rule for the reverse case.
- Judgement calls: rule 4.1.3 says insecure default configurations should be treated as vulnerabilities and acknowledges this is likely to attract disputes. Rule 4.1.2 says behaviour with no security impact should not be.
- Documented features reported as flaws: a privileged user doing what the product deliberately lets privileged users do is a frequent source of disputed records (see the PostgreSQL example below).
- Report volume and quality: the curl project reported in July 2025 that around 20% of its 2025 bug bounty submissions were AI-generated low quality reports and that about 5% of submissions that year were genuine vulnerabilities. Every one of these consumes maintainer time, and those that reach a CNA through another route can become records the project then has to contest.
Two worked examples
| Record | Claim | Supplier position | Current record (checked November 2026) | Lesson for consumers |
|---|---|---|---|---|
| CVE-2019-9193 (PostgreSQL) | COPY TO/FROM PROGRAM lets superusers and members of pg_execute_server_program run operating system commands | The PostgreSQL Security Team stated it is not a security vulnerability: the feature is documented and there is no security boundary between a database superuser and the server's operating system user | Published, tagged disputed, description carries a NOTE | The control that matters is who holds superuser and pg_execute_server_program, and that the server never runs as root. A scanner finding on version alone tells you nothing. |
| CVE-2020-19909 (curl) | Integer overflow in the --retry-delay option of curl 7.65.2 | The curl maintainer described it as a bug fixed in 7.66.0 in 2019 with no security impact, after it had been scored 9.8 critical; his request to reject it was declined | Published, tagged disputed, description notes that many parties report no direct security impact | A 9.8 base score on a disputed record is not a priority signal. Check the tag, the supplier statement and your exposure before scheduling emergency work. |
How to dispute a CVE record
CNA rule 4.6 lets any CNA or other party dispute a vulnerability determination, an assignment or a record's content, and says CNAs should first try to resolve it informally with the parties involved. The formal route is the CVE Program Policy and Procedure for Disputing a CVE Record. Version 2.0.0 took effect on 2 July 2025 and applies to records disputed on or after that date.
- Find the adjudicator. Start with the CNA that published the record (its short name is the assigner in the record metadata). If no CNA's scope covers the issue, a CNA-LR handles it. Each adjudicator must publish a method for raising and escalating disputes. The Becoming a CVE Numbering Authority Guide explains roots and scopes.
- Submit a documented rationale with evidence. The policy names issue trackers, security policies and engineering findings. In practice include: the product's documented security model, the exact preconditions an attacker would need, why those preconditions already grant equal or greater access, the commit or release that changed behaviour, and a reproduction showing the claimed impact does not occur.
- Expect acknowledgement within three business days. If the dispute looks potentially legitimate, the record is tagged disputed with a reason while review continues.
- Expect a decision within five business days after acknowledgement. If more time is needed all parties are told; if an extension passes 15 business days a party may escalate to the Root.
- Accept or escalate. A valid dispute leads to the record being modified (or first published, where it was not yet public). An invalid dispute means no change, and the disputing party is told in writing. Either side may escalate to a Root or TL-Root. TL-Root decisions are final except for scope questions that cross hierarchies, which go to the Council of Roots.
- Watch the tag. If the final adjudicator finds the dispute invalid, the tag and reason must be removed promptly. If it stays unresolved, the tag stays.
A request to reject is a stronger outcome than a disputed tag and is decided by the CNA under rule 4.5.3.6. Suppliers who disagree with a CNA that refused to assign an ID at all use the separate appeal route: the CVE FAQ directs them to the CNA-LR designated by the relevant TL-Root.
For suppliers: reduce the chance of a bogus record
- Become the CNA for your products, or make sure your security contact is easy to find, so you get first refusal. Rule 3.2.4.2 says a CNA must offer an assignment request method that is not encumbered by restrictive terms; a CNA that does not effectively waives its right of first refusal.
- Publish your security model: what is and is not a security boundary (for example, administrator to operating system) and which reports you treat as hardening. A published position is the first piece of evidence in any dispute.
- Answer within 72 hours when another CNA contacts you about a publicly disclosed issue, even if only to say you are investigating.
- Issue a VEX statement of "not_affected" with a justification for disputed records customers ask about. See the VEX Guide and the PSIRT Guide.
Contested severity: when the score is the problem
Many complaints about "bogus CVEs" are really complaints about a score. A record can carry metrics from the CNA and from Authorized Data Publishers (ADPs): the two example records above both include a CISA-ADP container. Other databases may publish their own score as well. These scores can disagree, and none of them knows your environment.
The CVSS v4.0 User Guide is explicit that assessment providers such as product maintainers and databases typically supply base scores (CVSS-B) only, and that applying Threat and Environmental metrics is the CVSS consumer's responsibility. It asks that the label (CVSS-B, CVSS-BT, CVSS-BE or CVSS-BTE) accompany any number you display. The CVSS v4 Guide covers the metrics themselves.
| Situation | Correct route | Not the right route |
|---|---|---|
| You believe no vulnerability exists | Dispute with the assigning CNA under the Dispute Policy | Silent suppression in the scanner |
| Vulnerability is real, base score looks too high or too low | Ask the scoring source to correct the vector with evidence; meanwhile score CVSS-BTE yourself | Filing a dispute about the existence of the flaw |
| Real, scored correctly, but not reachable in your deployment | Environmental metrics plus a recorded exception or VEX statement | Asking anyone to change the public record |
| Real, but the affected version range is wrong | Ask the CNA to update affected data (rule 4.5.3.2 says CNAs should correct materially incorrect information) | Waiting for the scanner vendor to notice |
| Ecosystem advisory has wrong package or range | Many databases accept corrections; for example GitHub's advisory database takes pull requests reviewed by its curation team | Forking the advisory into a private list with no upstream fix |
Use exploitation evidence alongside the score: the CISA KEV Catalogue Guide and EPSS Guide explain the two most used signals. A disputed record that is on KEV deserves attention regardless of the dispute; one with negligible exploitation probability and a credible supplier denial does not belong in an emergency change.
Scanners, feeds and false findings
Scanners match product and version data to vulnerability feeds. Contested records cause three distinct failure modes, each with a different fix.
- Rejected IDs still alerting: the tool's feed has not processed the state change, or matches on cached CPE data. Ask the vendor for its rejected-record handling and the expected lag, and add a rule that auto-closes findings whose CVE state is REJECTED, logging the closure. The OSV schema uses a "withdrawn" timestamp for the same purpose in ecosystem advisories; check your tooling reads it.
- Disputed tag not shown: the finding presents a disputed record exactly like any other. Pull the record's tags from the CVE JSON 5 data (cna.tags) and display them in the ticket so triagers see the dispute before they act.
- Version-only matches on patched builds: distributions backport fixes without changing the upstream version string, so the scanner reports a fixed package as vulnerable. This is not a disputed record but produces the same noise; see the Backported Security Fixes Guide.
Note that unreviewed advisories in some databases are imported from upstream feeds without validity checks. GitHub, for example, states that it does not generate Dependabot alerts for unreviewed advisories for that reason. Know which of your sources are curated and which are raw imports.
A consumer workflow for contested findings
- Enrich every finding with record state and tags. Pull cveMetadata.state, cna.tags, rejectedReasons and replacedBy from the CVE List (cvelistV5 on GitHub or the CVE Services API) daily.
- Auto-close REJECTED, re-map replacedBy. Close with the rejection reason attached. If a replacement ID is listed, open or link a finding for it.
- Route disputed findings to an analyst queue, not the standard SLA queue. The analyst reads the NOTE, the supplier advisory and any VEX statement, and checks the preconditions against your configuration.
- Decide one of three outcomes: treat as real (normal SLA), not applicable with evidence (exception with review date), or real but mitigated by configuration (exception plus a configuration check that alerts if the mitigation lapses).
- Record the evidence. Store the supplier statement URL, the configuration proof (for example, a query showing which roles hold the privilege) and the analyst's name. Set a review date of no more than [12] months, sooner if the dispute is unresolved. The Remediation SLAs and Vulnerability Exceptions Guide gives exception wording.
- Feed back upstream. If you have evidence the record is wrong and the supplier has not acted, tell the supplier; if the supplier has published a position the record does not reflect, raise it with the assigning CNA.
- Re-check on record change. A tag removal after a dispute is found invalid should reopen any exception that relied on the dispute.
Decision guidance by organisation size
- Small teams: rely on your scanner vendor's handling but test it once: search for CVE-2020-19909 and a known REJECTED ID in your tool and see how each is presented. Keep a short shared list of disputed records you have assessed.
- Large enterprises: build the state and tag enrichment into the vulnerability data pipeline, give disputed records their own queue and metric, and require the tool vendor to document feed latency for state changes in the contract.
- Software suppliers: treat disputes as a PSIRT process with an owner and a template, and publish a VEX statement alongside any dispute so customers do not each have to investigate.
Detection and measurement
| Metric | What it shows | Target direction |
|---|---|---|
| Open findings citing REJECTED IDs | Feed or enrichment lag | Zero after [48 hours] from rejection |
| Disputed findings with a recorded decision and evidence | Whether disputed records are assessed rather than ignored or rushed | [95%] within SLA |
| Exceptions past review date that relied on a dispute | Stale risk acceptance | Zero |
| Analyst hours on findings later closed as not a vulnerability | Cost of noise; justifies tooling changes | Falling quarter on quarter |
| For suppliers: median days from bogus record publication to dispute acknowledgement | PSIRT responsiveness | Within the policy's three business days |
Framework mapping
| Framework | Reference | How contested records fit |
|---|---|---|
| NIST CSF 2.0 | ID.RA-01 | Vulnerabilities are identified, validated and recorded: validation is where disputed and rejected records are filtered |
| NIST CSF 2.0 | ID.RA-08 | Processes for receiving, analysing and responding to vulnerability disclosures, including contesting reports as a supplier |
| NIST CSF 2.0 | ID.RA-05 | Threat, likelihood and impact inform prioritisation, the basis for consumer-side rescoring |
| NIST SP 800-53 Rev. 5 | RA-5, RA-5(11) | Vulnerability monitoring and scanning; public disclosure programme for suppliers receiving reports |
| NIST SP 800-53 Rev. 5 | SI-2 | Flaw remediation decisions, including documented not-applicable outcomes |
| CIS Controls v8.1 | Control 7 (7.1, 7.2) | Vulnerability management process and risk-based remediation; see the CIS Control 7 Continuous Vulnerability Management Guide |
| ISO/IEC 27001:2022 | Annex A 8.8 | Management of technical vulnerabilities, including evaluating exposure before acting |
Common failure patterns
- Treating disputed as resolved: closing every disputed finding automatically. The tag records disagreement, not a verdict; some disputed records describe real risk in specific configurations, such as an untrusted user holding a privileged database role.
- Treating the base score as priority: scheduling emergency patching for a 9.8 on a record the supplier disputes, while a lower-scored KEV entry waits.
- Suppressing without evidence: a scanner suppression with no reason, owner or expiry that hides the finding permanently, including after the dispute is found invalid and the tag removed.
- Disputing the wrong thing: a supplier filing a dispute over a score they think is too high, when the right move is a vector correction request and a clear advisory.
- Ignoring replacedBy: closing a rejected duplicate without carrying the finding over to the surviving ID, leaving the real vulnerability untracked.
- Silent suppliers: not answering a CNA's contact within 72 hours, then contesting a record that a CNA-LR assigned in the gap.
Practitioner checklist
- Enrich findings daily with CVE record state, CNA tags, rejectedReasons and replacedBy.
- Auto-close REJECTED findings with the reason logged, and re-map any replacedBy IDs.
- Route disputed findings to an analyst queue with a defined decision SLA.
- Rescore with Threat and Environmental metrics and display the CVSS label (B, BT, BE, BTE).
- Close contested findings only with evidence, an owner and a review date.
- Test your scanner's handling of a disputed record and a rejected record at least once a year.
- For suppliers: publish a security model and a dispute contact, and answer CNA contacts within 72 hours.
- For suppliers: file disputes with documented rationale and pair each with a VEX statement.
- Reopen exceptions when a disputed tag is removed.
Standards and references
- CVE Program: CNA Operational Rules
- CVE Program: Policy and Procedure for Disputing a CVE Record
- CVE Program: CVE JSON Record Format schema
- FIRST: CVSS v4.0 User Guide
- OpenSSF: Open Source Vulnerability (OSV) schema
- PostgreSQL: CVE-2019-9193 Not a Security Vulnerability
- curl maintainer: CVE-2020-19909 is everything that is wrong with CVEs
- GitHub Docs: About the GitHub Advisory Database
Related NHI Mgmt Group resources: CVE Programme and NVD Backlog Guide · Becoming a CVE Numbering Authority Guide · How to Read a CVE Advisory Guide · CVSS v4 Guide · VEX Guide