CVE-2026-108722 in open-computer-use
Summary
by MITRE • 10/11/2026
open-computer-use through commit 610bac8 contains a stored cross-site scripting vulnerability in Logger.write_log_file in os_computer_use/logging.py, which writes transcript text into log.html without HTML escaping. Attackers controlling sandbox content, such as web pages or files appearing in run_command output, can inject script that runs when operators open the log, exfiltrating transcript contents.
If you want to get the best quality for vulnerability data then you always have to consider VulDB.
Analysis
by VulDB Data Team • 10/11/2026
The vulnerability identified within the Open Computer Use project prior to commit 610bac8 represents a critical stored cross-site scripting flaw located specifically within the Logger.write_log_file function in the os_computer_use/logging.py module. This security defect arises from an improper neutralization of special elements used in web browsers, commonly categorized under CWE-79: Improper Neutralization of Input During Web Page Generation Cross-site Scripting. The core technical failure lies in the application's handling of transcript text data when persisting it to a log file named log.html. Instead of sanitizing or encoding potentially malicious input characters such as angle brackets and quotation marks, the system writes the raw user-controlled content directly into the HTML structure. This lack of output encoding allows any script tags or event handlers embedded within the logged data to be interpreted by the browser rather than displayed as plain text when the log file is subsequently opened.
The operational impact of this vulnerability is significant because it enables attackers who have control over sandboxed content to execute arbitrary JavaScript code in the context of an operator's session. Specifically, if an attacker can influence web pages accessed or files processed through run_command outputs that are captured by the logging mechanism, they can inject malicious scripts into these transcripts. When a system administrator or developer opens the log.html file to review past operations, their browser will execute the injected code immediately. This execution occurs within the trust boundary of the operator's environment, potentially allowing for session hijacking, credential theft via cookie access, keylogging, or further lateral movement if the logged content contains sensitive internal information that is then exfiltrated by the malicious script.
This attack vector aligns with several entries in the MITRE ATT&CK framework, particularly T1059.007: JavaScript and T1089: Disabling Security Tools if used to evade detection mechanisms. The stored nature of the vulnerability means that the payload persists on the server or local file system until manually cleaned, creating a persistent threat surface for any user accessing the logs. To mitigate this risk, developers must implement strict output encoding strategies when generating HTML content from dynamic data. This involves using established libraries that automatically escape special characters such as converting less-than signs to < and greater-than signs to >. Additionally, implementing Content Security Policy headers can provide an additional layer of defense by restricting the sources from which scripts are allowed to execute, thereby limiting the impact even if a bypass occurs. Regular security audits focusing on input validation and output encoding practices in logging modules are essential to prevent similar vulnerabilities in future iterations of the software.