CISA KEV action plan for CVE-2026-90817 in REDCap
CISA's decision to add CVE-2026-90817 to the Known Exploited Vulnerabilities Catalog changes the scale of the response. The vulnerability is not waiting for a theoretical proof of concept. It has a public technical chain, a known exploitation determination, and a short remediation window for covered federal organizations.
Vulnerability overview
Securifera reported unauthenticated remote code execution in REDCap and the CVE record uses CVSS 3.1 score 9.8. The affected range starts at REDCap 13.3.0. Fixed releases are 16.0.49 LTS, 17.3.10 LTS, and 17.4.4 Standard Release. CISA lists the issue as known exploited, added it on November 18, 2026, and assigned a November 21, 2026 due date.
Those dates make ownership and evidence preservation immediate tasks. The response should not begin with a blind restart or a version-number scan that cannot see survey configuration.
Mechanism and exploitation conditions
The public survey handler decodes __passthru twice for an allowlist lookup while a later route restriction sees the less-decoded value. A double-encoded filename can therefore load the root dispatcher and pass an attacker-selected route. During initialization, REDCap calculates rights before the survey identity exists and treats an undefined SUPER_USER state as super-user access. Data Import rights become available without a REDCap login.
The attacker then provides fname to Data Import. The application accepts a PHP stream or local resource and uses it in a read-modify-write sequence. A generated filter chain and BOM cause controlled bytes to be written to a predictable BladeOne cache. A public Parcel route renders the template, and BladeOne includes the cache as PHP.
A valid public survey hash is required, but no account, project membership, session, role, or CSRF token is needed. Survey hashes are designed for public distribution and can appear in URLs, logs, search indexes, and shared documents.
Impact
The attacker can execute code with the web process's privileges. This may expose research records, uploaded files, credentials, and network access. The organization should evaluate both security and research-data impact, including whether regulated or participant data was present.
CISA's KEV entry confirms exploitation in the wild. It does not establish that every vulnerable deployment was attacked or that any particular record was changed or stolen. That distinction is important when reporting to leadership and when deciding how to preserve evidence.
Affected products and scope
Start with every REDCap endpoint, including partner installations and recovery copies. Flag 13.3.0 and later releases before the applicable fixed build. Mark public-survey status, internet reachability, authentication configuration, and ownership for each host. Assign a higher priority to systems that accept public survey traffic and process sensitive data.
Exposure context
ZoomEye Search app="REDCap" returned 22,014 product fingerprints. The CVE Filter vul.cve="CVE-2026-90817" returned 0. Exposure is 22,014, the larger selected count. This is not a confirmed vulnerable or compromised asset count, and it should not be used as a victim tally.
Remediation and mitigations
Use this sequence for each affected deployment:
- Assign an incident and system owner before making changes.
- Record the REDCap release, branch, public-survey setting, proxy rules, and backup status.
- Preserve web, application, proxy, PHP, and file-integrity evidence.
- Patch to 16.0.49 LTS, 17.3.10 LTS, 17.4.4 Standard Release, or a later maintained build.
- If patching is delayed, restrict or disable public surveys and block the root dispatcher, Data Import, and Parcel routes from survey contexts.
- Restart the relevant services and verify the running build.
- Review for suspicious
__passthru, routes,fnamevalues, cache writes, child processes, and data access. - Rotate secrets and credentials available to the web process when risk or evidence supports it.
- Confirm external exposure and containment from outside the host.
- Close the incident only after patch verification, evidence review, and a defined monitoring period.
The completion check should be technical: the build string is correct, the route and survey controls are active where patching is incomplete, logs are retained, and owners have reviewed findings. A clean scanner result is not enough because the vulnerability is reached through application behavior and a valid survey identifier.
Top comments (0)