OT DETECTION · CVE-2021-26828

The HMI login succeeded.
What happened next?

ScadaBR's upload flaw shows why successful authentication is the beginning of the detection story. Follow the evidence from a valid session to server-side execution.

Deception Check Research · · 6 minute read

8.8 HIGHCVSS v3.1
AuthenticatedA valid session is required
Already in KEVAdded December 3, 2025

A successful login tells you an account was accepted. It does not tell you what that account did next.

For a vulnerable ScadaBR HMI, the next action can be a file upload that becomes code running on the server. A login alert, a web-access log, and a host alert can each show part of that sequence. The useful investigation connects them.

WHY WE ARE REVISITING IT TODAY

New repository. Older exploited vulnerability.

Our October 10 review was prompted by a GitHub repository created October 9, 2026, described by its author as corrected Python code for CVE-2021-26828. Public exploit material existed in April 2021; CISA added the flaw to its Known Exploited Vulnerabilities catalog on December 3, 2025.

The new repository is the reason for this review, not evidence of a new outbreak or a newly discovered exploit. It is a useful prompt to check whether an older remediation decision was completed and whether monitoring follows what happens after login.

  1. 2021Public exploit material
  2. DEC 2025Added to CISA KEV
  3. OCT 9, 2026Repository posted
  4. OCT 10, 2026Our defensive review

When an upload becomes a program

ScadaBR is a Java-based supervisory control and human-machine interface application. It can give operators a browser view of industrial data and a way to interact with connected controllers. The CVE record identifies affected ScadaBR versions through 0.9.1 on Linux and 1.12.4 on Windows, with authenticated JSP upload and execution through view_edit.shtm.

The security boundary is between accepting a file as data and allowing the server to execute it. A JSP file placed in an executable web-application location can cross that boundary. Execution runs with the service's permissions; it is not automatically root or administrator access.

Changing a default password matters, but it does not repair the upload vulnerability. A stolen or otherwise misused authorized account can also satisfy the login requirement. The resulting access to industrial systems depends on the host's configured connections, permissions, and network reach.

Connect four stages: an authenticated session, an upload request, a request for the uploaded file, and evidence of execution. Preserve account and request context, check the webroot, and correlate process, file, or network activity. HTTP success alone does not prove execution.
A detection model for investigation. A request can fail; collect the evidence that shows how far the sequence actually progressed.

This sequence has been observed

In research published October 9, 2025, Forescout described a March 2025 attack on its HMI honeypot: the visitor logged in with default credentials, used CVE-2021-26828 to upload a Java web shell, and performed basic host enumeration. That is stronger evidence than the existence of an exploit repository: the upload path was used against a reachable system.

The same report's later TwoNet incident involved the separate CVE-2021-26829 XSS flaw and HMI-level tampering. Those are different observations. The upload case belongs to Forescout's research, not our fleet, and a honeypot interaction is not evidence that a real industrial process was damaged.

Give the responder a connected sequence

A rule that finds .jsp is a starting point. A useful alert explains the relationship between the account, the upload, the resulting path, and any subsequent activity.

01 / SESSION

Who authenticated?

Retain the account, source, time, and a safe correlation identifier. Compare the activity with approved engineering work and review newly created or changed accounts. Keep passwords and bearer tokens out of alert payloads.

02 / UPLOAD

What was submitted?

Review activity at view_edit.shtm with its method, filename, content type, and available upload audit data. A suspicious filename is a lead; verify what the server stored. Routine page access is not exploitation.

03 / FILE

Was new code reachable?

Correlate a subsequent request with a newly created or modified executable file in the web application's served directories. Compare with approved deployments. An HTTP 200 response alone does not establish code execution.

04 / CONSEQUENCE

What did the host do?

Use available host and network telemetry to investigate unexpected Java or Tomcat child processes, file changes, or outbound activity. Some web-shell actions run inside the JVM; the absence of a child process does not clear the session.

Carry related activity into one investigation with original timestamps and evidence references. Keep attempted upload, confirmed file placement, and corroborated execution distinguishable. If host telemetry is unavailable, identify the gap and preserve what the application and network recorded.

Verify the remediation for the software you actually run

ScadaBR distributions and Scada-LTS releases are not interchangeable. CISA's entry points to Scada-LTS correction PR #2174; the associated fix appears in the 2.7.1 release history. That documents a correction in that project, not a universal fixed version for every ScadaBR package.

There is a further reason to check the exact build: DREAM Research Labs reported four separate ScadaBR 1.2.0 vulnerabilities in May 2026 and said no patch was available at publication. Those are different CVEs, including an authentication weakness and command injection. Resolving this older upload flaw is not a complete security assessment of that release.

  1. Reduce exposure with the operations owner. Remove direct public access to the HMI, preserve essential operator access, and restrict management paths to authorized systems and users. Coordinate changes with the people responsible for safe plant operation.
  2. Identify and verify. Record the distribution, build, servlet container, service account, and prior remediation. Check the actual correction and test the supported upgrade path in a representative environment. A higher version number by itself is not enough.
  3. Review access and retained evidence. Replace default credentials, audit accounts, and investigate suspicious uploads, executable files, and subsequent activity. Preserve relevant logs and file evidence before cleanup or replacement.
  4. Resolve unsupported systems deliberately. Where a supported, verifiable remediation is unavailable, use documented isolation and monitoring while planning migration or discontinuation. Do not run a public exploit against a production HMI to see whether it works.

TURN THE LESSON INTO A DETECTION TEST

Can your team follow the session?

A decoy HMI gives a team a place to test that question without granting access to the real process. In an approved, isolated design, it can record attempted interaction with a simulated login and upload surface while forwarding evidence to the responder.

Emulate the dangerous action; do not execute uploaded code. Keep the decoy outside process-control authority, constrain outbound access, and retain evidence off the decoy. Use a benign commissioning test to confirm that the alert arrives with enough context to investigate.

Our goal is an earlier warning that survives handoff: what was touched, what was requested, what the decoy returned, and who owns the response. That is a practical starting point for an OT deception pilot.

Discuss an OT deception pilot

Research notes

Sources and repository metadata checked October 10, 2026. The October 9 repository describes corrected code; we did not download or execute its exploit or establish a new exploitation capability. The KEV addition is historical, not an October 2026 addition. CISA records ransomware-campaign use as Unknown.

The observed upload incident is attributed to Forescout. No confirmed Deception Check capture of CVE-2021-26828 is claimed. The diagram and investigation steps are defensive guidance, not a replay of our telemetry. Product-specific remediation must be checked against the installed distribution and build.

Primary references: CVE record; CISA KEV entry; CISA-maintained catalog data; and the research and project records linked above.