I Had Claude MCP Hack Me
Source: YouTube · John Hammond · published Nov 14, 2025 · 21:09
The video demonstrates how Claude AI can be used as an "advanced persistent threat" through the Atomic Red Team MCP server to execute cybersecurity tests, highlighting both the potential and risks of AI in hacking scenarios 0:18-0:40.
Key Takeaways:
• The Atomic Red Team MCP server allows AI assistants like Claude to directly execute cybersecurity tests that emulate hacker behaviors, automating what was previously a manual process 1:48-2:18.
• The presenter demonstrates Claude executing various attack techniques like changing desktop wallpaper, creating registry run keys, and creating scheduled tasks 13:17-14:48.
• Security concerns are raised about AI having the ability to execute commands, with a recommendation for MCP gateways that provide visibility and control over AI activities 4:33-5:14.
• The tool has practical applications for cybersecurity professionals doing threat intelligence and detection engineering, potentially saving significant time 1:03-1:14.
This demonstration showcases AI's potential to transform cybersecurity testing while highlighting important security considerations when granting AI systems execution capabilities.
Sources:
- 0:18-0:40 Introduction to Claude as an advanced persistent threat
- 1:03-1:14 Manual cybersecurity work being time-consuming
- 1:48-2:18 Explanation of MCP server capabilities
- 4:33-5:14 Security concerns and MCP gateway solution
- 13:17-14:48 Demonstration of Claude executing attack techn
Generate CPE Credits
Generate a professional CPE document from this video's transcript.
Estimated credit: 0.5 CPE hours
Estimate uses the video runtime (1 hour ≈ 1 CPE, rounded to the nearest 0.5, minimum 0.5, maximum 2.0). The final amount can be lower after review, never higher.
Topic: Cybersecurity. Commonly maps to: Security and Risk Management, Security Operations. Exact CISSP domains are assigned during generation.
Free account. One generation at a time, with a daily limit.
CPEBuddy is independent and not affiliated with or endorsed by ISC2, ISACA, or any certification body. Exports are formatted for common CPE submissions; acceptance is at your certification body's discretion.
Transcript Preview
First 800 characters of the transcript
AI, Chat, GPT, Claude, MCP, and other AI acronyms and buzzwords. It's no secret artificial intelligence is the new hotness, and a lot of people have different opinions, right? Good, bad, ugly, I don't know. But I think it is a really interesting conversation when we talk about it in the lens of cyber security. Could there ever be an AI powered hacker or threat actor? Or could we have AI hack you or me? So, I thought, you know what? Maybe a cool video would be chatting a little bit about and getting a chance to showcase the atomic red team MCP when Claude becomes the AP or the advanced persistent threat. Now, as always, kudos. Credit where credit is due. Big thanks to Cyberbuff. This is his tool and blog post. He has been breaking things with an atomic red team MCP server because manual exp…