Reasonably Optimistic The temperature is rising on AI. What comes next?
Apr 15, 2026
Damir Marusic, Op-ed editor at The Washington Post who covers foreign policy and tech. He breaks down Anthropic’s Mythos and why models that find hidden software bugs are alarming. They talk zero-day vulnerabilities, Pentagon tensions with AI firms, protests over data centers, and what it takes to harden systems and stay competitive with China.
AI Snips
Chapters
Transcript
Episode notes
AI Development Has Arms Race Dynamics
- AI development has become an international arms race with China as a major competitor.
- That competition pressures rapid advancement and deployment even when risks remain unresolved.
Mythos Found A 27 Year Undiscovered Bug
- Anthropic's Mythos rapidly identified a 27-year-undetected bug in a Unix-like open source OS.
- That example shows even mature open-source projects can hide flaws until an AI with extensive search time uncovers them.
Zero Days Create Strategic Hold And Use Risks
- Zero-day vulnerabilities give adversaries a hold-it-and-wait advantage, turning undisclosed flaws into strategic weapons.
- States or criminals can exploit undisclosed bugs until defenders patch systems, making discovery timing critical.