AiCyber.Land #23 - Is Your AI Hiding Secret Messages From You?

Ever tried to get a McDonald's AI to write Python code for you? Or wondered if you could hide secret messages inside AI-generated text? In this episode of AI and Cyberland, we're diving deep into the wild world of AI attacks and cyber shenanigans. From Google's latest research on prompt injection to a new tool that turns LLMs into spy gadgets, you won't want to miss this. --- IN THIS EPISODE: Welcome back to the AI and Cyberland podcast! This week, Bryce and Shelby get briefed on the latest in AI and cybersecurity, and things get weird... and hilarious. First up, Shelby breaks down Google's new research on prompt injection attacks. While the number of attacks is rising, their sophistication is surprisingly... lame. We explore the difference between direct and indirect attacks, the kinds of traps hackers are setting for AIs in the wild (from pranks to actual data theft), and what Google's research uncovered about exfiltration and destruction attempts. Plus, we share some truly amusing stories of people jailbreaking customer service bots at McDonald's and Chipotle to get them to write code! Then, Bryce introduces a fascinating proof-of-concept tool called "Tomato" that enables steganography in LLM outputs. That's right—hiding encrypted, invisible messages inside seemingly random AI-generated text. It's like writing with invisible ink for the modern age, and we brainstorm a brilliant (and definitely not legal advice) use case for tech executives. This sparks a hilarious discussion about the AI bubble, featuring the shoe company Allbirds pivoting from wool shoes to... AI. Finally, we get into some modern-day vigilante justice. Shelby highlights the work of Jake Swizz, who is "war driving" and exposing the locations and vulnerabilities of the notoriously invasive Flock security cameras. We rant about the company's terrible security practices and why everyone should be concerned about their reach. --- KEY MOMENTS: ⏱️ **KEY MOMENTS:** 00:41 - Google's Latest Research on AI Prompt Injection 05:32 - Can The McDonald's AI Bot Write You a Python Script? 09:20 - Hiding Secret Encrypted Messages in LLM-Generated Text 11:33 - A Shoe Company Pivots To "AI" (The Bubble is Real) 17:17 - Vigilante Hacker "War Drives" Flock Surveillance Cameras 21:45 - What We're Watching: Game of Thrones Prequel Review --- CONNECT WITH US: What's the craziest thing you've seen an AI do? Have you ever tried to trick a customer service bot? Drop your stories in the comments below! If you enjoy our deep dives into the intersection of AI and Cyber, make sure to hit that LIKE button, SUBSCRIBE to the channel, and turn on notifications so you never miss a brief. #AISecurity #PromptInjection #CyberSecurity #LLM #Steganography #FlockCameras #Hacking #TechPodcast