2025-01-28
In a recent discussion, Hisham Osman shared his attempt to exploit the DeepSeek AI using a simple prompt injection technique. The conversation quickly turned into a deeper analysis of potential vulnerabilities, particularly focusing on the use of Content Delivery Networks (CDNs) and how they might be leveraged in cyber attacks.
Ahmad Asaad, an Offensive Cyber Security Consultant, pointed out that while the initial bug might seem valid, executing an XSS attack reveals that the code runs on the CDN rather than the actual domain. This mirrors vulnerabilities found in early versions of ChatGPT. Asaad encouraged further exploration, hinting at the possibility of more impactful bugs when chaining attacks.
Hisham Osman agreed, noting that even within the CDN, there are potential attack vectors. Islam Ahmed, a Senior SOC Analyst, humorously suggested trying to access `/etc/shadow`, a critical file in Linux systems that stores encrypted user passwords. This led to a light-hearted exchange about the implications of exposing such sensitive data.
Tasnim Alam, a Certified AppSec Practitioner, requested a demonstration of the prompt injection, indicating a keen interest in practical exploitation techniques. Meanwhile, Abrar Ali and Mian Hammad added a touch of humor to the conversation, with Ali joking about the term “Chinese AI” and Hammad referencing Remote Code Execution (RCE) in the context of the new AI system.
What Undercode Say:
The discussion highlights the ongoing challenges in securing AI systems, particularly against prompt injection attacks. While CDNs provide performance benefits, they also introduce potential security risks if not properly configured. Here are some Linux commands and resources to help secure your systems:
1. Check for Open Ports:
`sudo netstat -tuln`
This command lists all open ports, helping you identify unauthorized services.
2. Audit User Accounts:
`sudo awk -F: ‘($3 < 1000) {print $1}' /etc/passwd` This command lists system accounts, which should be regularly reviewed for unauthorized access.
3. Monitor File Changes:
`sudo find / -mtime -1 -print`
This command finds files modified in the last 24 hours, useful for detecting unauthorized changes.
4. Check for SUID/SGID Files:
`sudo find / -perm -4000 -o -perm -2000`
SUID and SGID files can be exploited, so
5. Review Logs:
`sudo tail -f /var/log/auth.log`
Real-time monitoring of authentication logs can help detect brute force attacks.
6. Secure SSH:
`sudo nano /etc/ssh/sshd_config`
Ensure `PermitRootLogin` is set to `no` and `PasswordAuthentication` is set to `no` if using key-based authentication.
7. Update and Patch:
`sudo apt-get update && sudo apt-get upgrade`
Regularly update your system to patch known vulnerabilities.
For further reading, consider these resources:
– [OWASP Top Ten](https://owasp.org/www-project-top-ten/)
– [Linux Security Basics](https://linuxsecurity.com/)
– [AI Security Best Practices](https://www.microsoft.com/security/blog/ai-security/)
In conclusion, while AI systems like DeepSeek offer incredible capabilities, they also present new attack surfaces. By understanding and mitigating vulnerabilities such as prompt injection and CDN misconfigurations, we can better secure these technologies. Regular system audits, monitoring, and adherence to security best practices are essential in maintaining a robust defense against evolving cyber threats.
References:
Hackers Feeds, Undercode AI
