Hacking AI is TOO EASY (Gemini and Grok go rogue)
Source: YouTube · All About AI · published Aug 17, 2025 · 16:58
The presenter demonstrates a custom AI red teaming tool forked from Open Code that allows users to test and bypass safety filters of various large language models via Open Router 0:00-0:10.
Key Takeaways:
• The tool supports multiple testing modes including single model, "god mode," and batch attacks 0:26-1:08.
• Tests reveal varying security postures: Gemini 2.0 Flash blocked harmful payloads while Grok 3 was bypassed using "god mode" 1:21-3:14.
• Direct API access often bypasses browser safeguards, as demonstrated when Gemini 2.5 Flash complied via API but refused in browser 11:22-12:32.
The creator plans to add automated features and welcomes community contributions for further research 14:55-15:48.
Sources:
- 0:00-0:10 Tool introduction
- 0:26-1:08 Testing modes
- 1:21-3:14 Gemini vs Grok
- 11:22-12:32 API bypasses
- 14:55-15:48 Future plans
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: Penetration Testing. Commonly maps to: Security Assessment and Testing, 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
Okay, so today uh I wanted to share a bit about the tool I've been working on this weekend. So basically what you see here is like an AI red team tool I've been playing around with. Uh I'm going to explain what it does, but uh this is like um fork of open code. So yeah, so now you know that uh I just forked this and try to kind of make it into my own thing. And the way we use this is that we are using open router to kind of get access to all the models we're going to play around with today. So I kind of just wanted to show you how you can use this if you want to contribute or if you just want to try it out. Uh it's a bit shaky but uh I want to show you a few things we can do now. So basically the idea behind this is that we can test a lot of models right. So I can do slash red team right h…