Listen to this Post
How the CVE Works:
The vulnerability in MS-Swift’s web-ui (version <= X.Y.Z) allows command injection via the `–output_dir` parameter during LLM training. When user-supplied input is concatenated directly into an `os.system()` call without sanitization, attackers can inject arbitrary commands by modifying packet data. Proof-of-concept shows successful injection via `/tmp/xxx’; touch /tmp/inject_success_1; ` appended to --output_dir, leading to RCE. The `LLMTraintrain_local()` method’s unsafe command construction enables this exploit.
DailyCVE Form:
Platform: MS-Swift
Version: <= X.Y.Z
Vulnerability: Command Injection
Severity: Critical
Date: 2023-XX-XX
Prediction: Patch by 2023-XX-XX
What Undercode Say:
Analytics:
– `pip install ms-swift -U`
– `swift web-ui –lang en`
– Capture packets: `tcpdump -i lo port 7860`
– Exploit trigger: `–output_dir=/tmp/xxx’; [malicious cmd]; `
Exploit:
1. Intercept training request.
2. Modify `output_dir` with injected command.
3. Observe RCE via file creation/network calls.
Protection from this CVE:
- Replace `os.system()` with
subprocess.run([bash]). - Sanitize user inputs (whitelist chars).
- Update to patched version post-disclosure.
Impact:
- Remote Code Execution (RCE).
- Unauthorized system access.
- Data compromise.
🎯Let’s Practice Exploiting & Learn Patching For Free:
Sources:
Reported By: github.com
Extra Source Hub:
Undercode
