

Episode 123: Hacking AI Series: Vulnus ex Machina - Part 2
May 22, 2025
The discussion dives into mastering prompt injection, showcasing how emotional tactics can manipulate AI models. Key AI vulnerabilities are explored, with insights into bug hunting tools and recent leaks. The comparison between Google's AI technologies highlights advanced exploitation techniques. Emphasizing meticulous testing, the hosts share strategies for recognizing and categorizing vulnerabilities. The episode wraps up with a sneak peek into future content on emerging risks within AI exploitation, encouraging listeners to engage on social media.
AI Snips
Chapters
Transcript
Episode notes
Use grep.app for Efficient Hunting
- Use grep.app for fast searches across many GitHub repos to find potential bug bounty targets.
- The new Next.js-based grep.app is much faster and indexes more repositories.
O3 AI Demonstrates Agentic Reasoning
- O3 AI is highly advanced with agentic style reasoning surpassing most models.
- It helps solve complex problems by iterative tool use and research steps.
Frame Delivery and Impact
- When attacking AI, focus on delivery and impactful actions of prompt injections.
- Use these steps to frame severity and exploit vulnerabilities effectively.