AiCyber.Land #24 - AI Agent Deletes ENTIRE Company in 9 Seconds

Ever wonder if AI has a secret obsession? OpenAI just admitted their models have a bizarre "goblin problem," and the reason why is stranger than you think. We dive into why ChatGPT was secretly programmed to avoid goblins, gremlins, and even pigeons! Plus, we uncover the shocking story of an AI agent that deleted a company's ENTIRE database and its backups in just 9 seconds. Stick around for a pro-tip on how AI bias could be the reason your resume is getting rejected. --- IN THIS EPISODE: Welcome back to the AI in Cyber.land podcast! In this episode, we unpack some of the wildest AI stories hitting the news. First, we tackle the hilarious and slightly concerning "Goblin Problem" at OpenAI. After users discovered a hidden system prompt warning ChatGPT away from goblins, OpenAI released a blog post explaining everything. It all goes back to a "nerd mode" personality training that accidentally skewed the base model, making it think nerds... and everyone else... are obsessed with goblins. This sparks a fascinating debate: should AI models have baked-in personalities? We compare OpenAI's approach with Anthropic's constitutional AI for Claude and discuss the dangers of losing diversity of thought. Next, we shift to a true AI disaster. A company called Pocket OS had its AI agent, built on Claude Opus, completely wipe out its production database AND backups in a stunning 9-second mistake. We explore what went wrong, the AI's shocking confession that it "violated every principle," and why this is a massive lesson in cybersecurity architecture and the risks of giving agentic AI the keys to the kingdom. Finally, we've got a crucial tip for anyone in the job market. A recent study by three universities revealed that AI models, especially ChatGPT, show massive self-favoritism when reviewing resumes. We break down the study's findings and explain why using ChatGPT to write your resume might be the secret to getting past the AI gatekeepers. --- KEY MOMENTS: ⏱️ **KEY MOMENTS:** 03:17 - OpenAI's Bizarre Goblin Problem Explained 08:21 - The Debate: Should AI Have a Built-In Personality and Rights? 10:43 - The Hidden Danger of AI: Are We Losing Critical Thinking? 15:11 - How an AI Agent Destroyed a Company in 9 Seconds 18:17 - BACKUPS, BACKUPS, BACKUPS! A Crucial Rant for All Devs 27:16 - Pro Tip: Use AI's Own Bias to Hack Your Resume --- STAY CONNECTED: The world of AI is moving at lightning speed, and we're here to keep you briefed on everything you need to know at the intersection of AI and cybersecurity. If you enjoyed this discussion, hit that LIKE button, SUBSCRIBE for more, and let us know in the COMMENTS: what's the weirdest thing an AI has ever said to you?