Podcast
A.I. Safety Is So Back + Mythos Mayhem With Nikesh Arora + Hot Mess Express
Hard Fork
- Mythos Forced A Fast Shift In US AI Policy
- The Trump administration rapidly shifted from dismissing AI safety to considering pre-release model vetting after encountering a powerful model like Claude Mythos.
- Mythos’s ability to find novel code vulnerabilities forced policymakers to treat frontier models as a distinct national-security risk rather than ordinary technology. (Time 0:03:34)
- Turf Fight Over Who Reviews Frontier Models
- There is a turf war inside the administration over who should vet models, with Commerce’s Casey group and intelligence agencies both staking claims.
- That organizational confusion matters because locating review authority affects transparency, technical capability, and whether vetting favors national security or industry priorities. (Time 0:06:38)
- AI Shrinks Dwell Time To Minutes
- AI has shrunk the attacker dwell time from days to minutes, forcing defenses to operate at minute-or-second speed.
- Existing cybersecurity infrastructure was built for days-long response windows and must be overhauled to be ‘AI ready’ to fight AI with AI. (Time 0:29:22)
- Use Temporary Signatures While Patching Vulnerabilities
- Patch quickly and deploy temporary perimeter signatures to block exploit attempts while organizations fix vulnerable code.
- Palo Alto used signatures as scaffolding to buy time after Mythos-style scans found many daisy-chained vulnerabilities. (Time 0:30:30)
- Palo Alto’s Mythos Audit Found Seven Times More Bugs
- Palo Alto ran a concerted internal audit with hundreds of engineers using Mythos and found seven times more vulnerabilities than in a normal period.
- The exercise felt like a ‘cleansing’ of long‑standing tech debt and vulnerability backlog. (Time 0:31:00)