CVE-2026-12047: HTML Injection in pgAdmin 4 Cloud Deployment Wizard

Technical summary

The vulnerability exists across multiple cloud-integration endpoints (/rds/verify_credentials, /azure/check_cluster_name_availability, /google/verification_ack, /deploy, /update_cloud_server, and related routes) that propagate unescaped exception text from AWS, Azure, and Google SDKs into JSON response bodies (info and errormsg fields). The Cloud Wizard frontend uses html-react-parser to render these strings, enabling arbitrary HTML/iframe injection. The primary attack path involves crafting an access_key parameter containing structural HTML; when the cloud provider's STS service rejects the credential with an IncompleteSignature exception that quotes the access_key, pgAdmin forwards the exception text verbatim to the frontend. The injected iframe executes cross-origin JavaScript capable of modifying parent.location. The fix introduces a sanitize_external_text helper in web/pgadmin/utils/text_sanitize.py that HTML-escapes and strips control characters from all external/SDK exception strings before they enter JSON responses, and updates the FormFooterMessage frontend component to render backend-derived strings in plain-text mode to prevent HTML parsing even if a future code path forgets to escape.

Business impact

The immediate risk is low-severity: an authenticated internal user supplies a malicious credential, sees their own session redirected to an attacker-controlled site, and may be phished for credentials or infected with malware. However, the vulnerability's presence across multiple endpoints and the availability of a CSRF amplification path means an attacker with network access (e.g., a compromised internal system, malicious advertisement, or supply-chain compromise) could target pgAdmin deployments with crafted requests that redirect legitimate administrators during cloud deployment workflows. Organizations using pgAdmin to manage production cloud infrastructure may be disrupted or deceived into granting attacker-controlled infrastructure permissions.

Affected systems

pgAdmin 4 versions 6.6 through 9.15 are affected. The vulnerability is present in any deployment that exposes the cloud wizard module (AWS RDS, Azure, or Google Cloud integration). pgAdmin installations running only local/non-cloud database management are unaffected unless they have upgraded to include the vulnerable cloud blueprint code.

Exploitability

Exploitability requires authentication (a valid pgAdmin login) and user interaction (the victim must attempt to add or verify cloud credentials). The attacker can self-target by submitting a malicious credential request, immediately seeing the injection in their own session. Attacking other authenticated users requires an additional CSRF primitive to submit the malformed credential request with a valid X-pgA-CSRFToken, which raises the bar but remains feasible in scenarios where pgAdmin is accessed from untrusted networks or malicious ads reach administrator browsers. No known public exploit exists (CVE-2026-12047 is not tracked in the CISA Known Exploited Vulnerabilities catalog). The attack surface is narrow: only users interacting with cloud deployment workflows during credential verification are at risk in-session.

Remediation

Update pgAdmin 4 to version 9.16 or later. The patch applies HTML-escaping and control-character stripping to all cloud provider exception text via the new sanitize_external_text utility, and ensures the frontend FormFooterMessage component renders backend-derived strings as plain text. Verify against the vendor advisory that your specific patch version contains the updated code in web/pgadmin/utils/text_sanitize.py and the FormFooterMessage changes.

Patch guidance

Apply pgAdmin 4 version 9.16 or later as soon as practical. The fix is contained in a single merged utility module and frontend component update, making it low-risk to deploy. Test the patched version in a staging environment to confirm cloud credential workflows (AWS, Azure, Google) function normally. If you cannot immediately patch, disable cloud deployment features in pgAdmin's configuration until an update is available; this removes the attack surface entirely.

Detection guidance

Monitor pgAdmin access logs and proxy logs for suspicious /rds/, /azure/, /google/, and /cloud/ API requests containing HTML/iframe patterns in the access_key, database_name, or other credential parameters (e.g., strings like '<iframe', 'src=', angle brackets). Check pgAdmin's application logs for exception messages that appear unusual or contain HTML fragments. If pgAdmin is exposed to untrusted networks, monitor for abnormal redirect or page-load behaviors in user sessions during cloud credential submission. Look for repeated 400/403 responses from cloud provider backends (which often accompany the injection attack pattern), followed by unusual JSON error messages containing HTML.

Why prioritize this

Although the CVSS score is low (3.5), this vulnerability warrants prompt attention because it affects authenticated users during critical infrastructure-deployment workflows. An administrator attempting to provision cloud databases for production could be redirected to a phishing site or malware distribution point, potentially compromising their credentials or the entire cloud account. The fix is mature, low-risk, and available in a released version. Organizations using pgAdmin's cloud integration should prioritize patching; those not using cloud features can defer, but should still track the update as a future security baseline upgrade.

Risk score, explained

CVSS 3.5 reflects low immediate severity: the attack requires authentication (L in PR:L), user interaction (R in UI:R), and targets only the integrity of the current user's session (I:L for Integrity impact, no Confidentiality or Availability loss). The scope is Unchanged (S:U), meaning the injection is confined to pgAdmin's own DOM and does not cross trust boundaries in the strict CVSS sense. However, the practical risk is higher than the score suggests because successful exploitation during cloud infrastructure provisioning can lead to account takeover or deployment of attacker-controlled infrastructure. Defenders should weight this as a medium-priority patch due to context (cloud-admin workflows), not the raw CVSS number.

Frequently asked questions