CVE Analysis 3 min read

CVE-2026-6603: A vulnerability was determined Code injection

CVE-2026-6603 CVSS 7.3 HIGH vulnerability analysis

CVE-2026-6603 is a high severity vulnerability (CVSS 7.3) identified in the National Vulnerability Database. A vulnerability was determined in modelscope agentscope up to 1.0.18. Affected by this vulnerability is the function execute_python_code/execute_shell_command of the file src/AgentScope/tool/_coding/_python.py. This manipulation causes code injection. The attack is possible to be carried out remotely. The exploit has been publicly disclosed and may be utilized.

TL;DR: This is a code injection vulnerability (CVSS 7.3 HIGH). Authentication: Unknown. Patch immediately.

What This Vulnerability Is

A vulnerability was determined in modelscope agentscope up to 1.0.18. Affected by this vulnerability is the function execute_python_code/execute_shell_command of the file src/AgentScope/tool/_coding/_python.py. This manipulation causes code injection. The attack is possible to be carried out remotely. The exploit has been publicly disclosed and may be utilized. The vendor was contacted early about this disclosure but did not respond in any way.

The National Vulnerability Database assigned this issue a CVSS base score of 7.3, placing it in the HIGH category. The identifier CVE-2026-6603 was published on 2026-04-20T05:16:15.353.

CVE Identifier
CVE-2026-6603
CVSS Base Score
7.3 / 10.0 (HIGH)
Published
2026-04-20T05:16:15.353
NVD Entry
https://nvd.nist.gov/vuln/detail/CVE-2026-6603
MITRE Entry
https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2026-6603

Who Is Affected

Based on the vulnerability data published by NVD, the following products and configurations are identified as affected:

  • Specific affected products are listed in the NVD entry. Review the CPE data for your stack.

If your organization uses any of the above, this vulnerability applies to your environment. Even if your specific version is not listed, review the full CPE match data to confirm.

What Attack Traffic Looks Like

Illustrative patterns for defensive detection. Not a working exploit.

Code injection attacks insert executable statements into application input that gets interpreted at runtime. Watch for:

eval("user_input")
exec(compile(user_data, '', 'exec'))
new Function(user_input)()
${runtime_expression}
__import__('os').system('id')

How to Detect It

Sigma Rule

title: Code Injection Patterns in Application Logs
id: auto-cwe94-detection
status: experimental
description: Detects dynamic code execution patterns in application logs indicative of code injection.
logsource:
  category: application
detection:
  selection:
    message|re: '(eval\(|exec\(|compile\(|Function\(|__import__|Runtime\.getRuntime)'
  condition: selection
level: high

IOC Hunt List

  • Application logs showing eval/exec/compile function calls with user-supplied arguments
  • Unexpected child processes spawned by the application runtime
  • New files created by the application process outside normal paths
  • Outbound connections from application server to unknown destinations

What to Do About It

Here is what we recommend, in order of priority:

  1. Check your exposure. Determine whether the affected software or component is present in your environment. Asset inventories and software composition analysis (SCA) tools are the fastest route.
  2. Apply the patch. If a vendor patch or updated version is available, apply it. Check the references below for vendor advisories.
  3. Mitigate if patching is not immediate. If you cannot patch right now, evaluate whether network segmentation, access control changes, or configuration adjustments reduce the attack surface for this specific vulnerability.
  4. Monitor for exploitation. Check whether proof-of-concept exploit code exists. The CISA Known Exploited Vulnerabilities catalog tracks actively exploited CVEs.
  5. Document your response. Record what you checked, when you patched, and what residual risk remains. This matters for compliance and for incident response if this vulnerability is exploited later.

Triage Quick Start

Run as administrator. Collect artifacts before applying patches or making changes:

# Linux/Unix systems
tar czf /tmp/triage-$(hostname)-$(date +%s).tgz \
  /var/log/syslog /var/log/auth.log /var/log/messages \
  /var/log/secure /var/log/apache2/ /var/log/nginx/ \
  /etc/passwd /etc/shadow \
  /root/.bash_history \
  /home/*/.bash_history \
  /tmp/ /var/tmp/ 2>/dev/null

# Windows systems (PowerShell)
# Compress-Archive -Path C:\Windows\System32\winevt\Logs\*,C:\inetpub\logs\* -DestinationPath C:\triage.zip

If you need help analyzing these artifacts, call Sherlock Forensics at 604.229.1994.

Why This Matters for Companies Without Security Teams

A CVSS score of 7.3 means this vulnerability is straightforward to exploit, likely to cause significant damage, or both. For startups and small companies operating without a dedicated security team, vulnerabilities at this severity level represent real operational risk rather than theoretical concern.

The challenge is not awareness. Vulnerability databases are public. The challenge is triage: understanding whether a given CVE affects your specific stack, and knowing what to do about it before an attacker does. If you lack the internal capacity to perform that assessment, an external review of your exposure is a concrete next step.

Sherlock Forensics provides vulnerability assessment and penetration testing for organizations that need to understand their attack surface without building a full internal security function.

References and Further Reading