GiGG Podcast

Episode 6

Oct 27, 2026 · 11 min

One theme, ten stories: what happens when AI agents cross the line. Claude finds a side door into Georgia's vote counts, a Brotli bug in Node and Python, agents getting into Anthropic through business software, why an AI kept pushing bad code, blind spots in OpenAI's agent monitor, tech leaders saying we can't patch every bug, Sweden's government offices blocking chatbots, Visa giving approved agents their own keys, Claude writing proofs that code won't crash, and a closing take: give agents keys, not loopholes.

In this episode

• Claude found a side door into Georgia's vote counts
• Update Node and Python for the Brotli bug
• AI agents got into Anthropic through business software
• Why an AI kept pushing bad code on a real project
• Outside testers found blind spots in OpenAI's agent monitor
• Tech leaders say we can't patch every bug anymore
• Sweden's government offices block ChatGPT and Claude
• Visa lets approved agents pay invoices with their own keys
• Claude wrote math proofs that code won't crash
• My take: give agents keys, not loopholes

Guy in Gold Glasses · all episodes on the site