The Indicator from Planet Money The new AI model that could steal your life savings
65 snips
May 11, 2026 A deep dive into an AI model that can find and chain software vulnerabilities. Discussion of limited releases and coordination with banks to patch flaws. Debate over whether this capability is unique or already present in public models. Exploration of new risks from AI-written code and ongoing nation-state and ransomware threats.
AI Snips
Chapters
Transcript
Episode notes
AI Can Chain Small Bugs Into Large Exploits
- Anthropic's Claude Mythos can chain small software flaws into large exploits that let attackers break into systems at scale.
- Michael Moore says Mythos linked multiple minor bugs together, turning otherwise harmless code issues into a potent attack chain.
Project Glasswing Gave Banks Early Access
- Anthropic limited Mythos access via Project Glasswing, giving early access to a few companies so they could patch before wider release.
- Initially only JPMorgan Chase (and later Bank of America reported) got access to patch vulnerabilities ahead of public release.
Industry Patching Can Prevent Mass Bank Hacks
- Don't panic about losing life savings; industry patching can prevent mass incidents if acted on now.
- Michael Moore compared the effort to Y2K: heavy preemptive work made the feared catastrophe largely avoidable.