CVE-2023-36584

What it actually does

This vulnerability bypasses Windows Mark of the Web protections, which normally tag files downloaded from untrusted locations so Windows, Microsoft Office, and other applications can apply additional warnings, sandboxing, or Protected View behavior. A crafted file can avoid or suppress those trust-boundary protections after a user opens it, reducing integrity and availability safeguards and making follow-on code execution or malicious document activity easier when paired with another payload or social-engineering lure.

Why it matters to the business

Although the base severity is moderate, this issue is in CISA KEV, has confirmed in-the-wild exploitation, and had a federal remediation deadline of 2023-12-07, making it operationally urgent for any organization still missing the relevant fixes. Successful exploitation can cause users to open malicious files with fewer warnings, increasing the likelihood of malware infection, credential theft, data exposure through follow-on tooling, and disruption from payloads delivered through phishing or web downloads. The provided affected-product metadata lists pymatgen packages and fixes, which does not align with the Windows Mark of the Web vulnerability description; teams should verify exposure against authoritative vendor advisories and their software inventory before closing remediation.

How an attack unfolds

An attacker prepares a specially crafted file, archive, or document intended to be treated as less risky than a normal internet-downloaded file. The attacker delivers it through phishing, messaging, a compromised website, or another user-driven download path. When the targeted user opens or extracts the content, Windows Mark of the Web protections may not be applied as expected, so downstream applications may not show the usual warning prompts or may not enter Protected View or equivalent restricted handling. The attacker then relies on embedded content, macros, scripts, shortcut behavior, or a chained vulnerability to execute additional malicious actions. Because the bypass weakens a security warning and policy layer rather than directly granting code execution by itself, exploitation is commonly part of a broader social-engineering and malware-delivery chain.

How to fix it

Apply the vendor security update that addresses the Windows Mark of the Web bypass on all supported Windows systems, prioritizing endpoints used for email, web browsing, document handling, development, and administrative access. Because this vulnerability is listed in CISA KEV and was observed in active exploitation, treat unpatched systems as high-priority even though the CVSS score is 5.4. Enforce defense-in-depth controls for downloaded content: keep Microsoft Office and document viewers updated, block or restrict macros from internet-origin files, use attachment sandboxing, disable automatic execution from archives where feasible, and ensure endpoint detection controls monitor Office, scripting, and archive extraction behavior. The supplied package-fix data lists pymatgen fixes, including Debian 12 pymatgen fixed in 2022.11.7+dfsg1-11+deb12u1 and materialsproject/pymatgen fixed in commit c231cbd3d5147ee920a37b6ee9dd236b376bcf5a; because that does not match the Windows MOTW vulnerability description, validate package remediation separately and do not treat those package versions as the Windows fix.

How to confirm the fix worked

Confirm that all in-scope Windows assets have the applicable Microsoft security update installed and that vulnerability scanners no longer report exposure for the Mark of the Web bypass. Validate patch state through endpoint management, OS build and cumulative update inventory, authenticated vulnerability scanning, and vendor advisory checks. For control validation, test with benign internet-downloaded files and archives to confirm that Zone.Identifier alternate data streams are preserved where expected and that Office, SmartScreen, browser, and endpoint controls still apply appropriate warnings or restrictions. Also verify that software inventory findings are not conflating this Windows vulnerability with unrelated pymatgen package advisories; reconcile scanner plugin output against vendor advisory references and installed product evidence.

Typical attack path

1. Attacker creates a crafted file, document, archive, or lure designed to avoid normal internet-origin trust marking.
2. Attacker delivers the content by email, chat, web download, or a compromised file-sharing location.
3. User downloads, extracts, or opens the content.
4. Windows or a dependent application fails to apply expected Mark of the Web-based restrictions or warnings.
5. The malicious content runs with fewer user-facing prompts or reduced sandboxing compared with a normally marked internet file.
6. Attacker executes follow-on malware, scripts, credential theft tooling, or additional exploits depending on the payload.

How to detect it

This bypass does not produce a unique, reliable host artifact by itself, so detection should focus on evidence of suspicious content delivery and post-open execution. Review email security logs, web proxy logs, browser download histories, EDR telemetry, Windows event logs, Office telemetry, archive extraction activity, and file system metadata for files obtained from external sources that were opened by users and followed by unusual child processes. Look for internet-sourced documents, shortcuts, scripts, disk images, or archives in Downloads, Temp, email attachment, and collaboration-tool cache locations that do not have an expected Zone.Identifier alternate data stream. Correlate file-open events with Office or document-viewer child processes such as PowerShell, cmd, wscript, cscript, mshta, rundll32, regsvr32, curl, certutil, bitsadmin, or installers. Investigate sequences where a browser, mail client, Teams/Slack-like collaboration app, or archive utility writes a file and a user-facing application opens it shortly before outbound network connections, credential access, persistence changes, or payload staging. Because confirmed exploitation exists, include historical hunting around the publication and patch window as well as any period where endpoints remained unpatched.

What your security team should do

• Prioritize patch verification across Windows endpoints because the issue is KEV-listed and actively exploited in the wild.
• Review vulnerability-management data for mismatched product attribution; confirm whether findings refer to Windows Mark of the Web or unrelated pymatgen package advisories before assigning remediation owners.
• Hunt for suspicious user-opened files from email, browser download, temporary, archive extraction, and collaboration-tool directories that lack expected Mark of the Web metadata.
• Increase monitoring for Office, browser, archive utility, and scripting-engine process chains, especially document viewers spawning PowerShell, Windows Script Host, command shells, mshta, rundll32, regsvr32, or installer processes.
• Maintain or strengthen macro restrictions, attachment detonation, web download reputation checks, and endpoint isolation procedures for systems showing suspicious post-open behavior.
• Educate users that malicious downloaded files may not always display expected warning banners and should still be treated cautiously when received unexpectedly.