Nate Soares: Glasswing missing a vuln for 6+ months indicts defense-over-offense thesis
So8res · x · 2026-10-22
- Nate Soares (MIRI) argues that if tools like Glasswing fail to find a vulnerability for 6+ months, that indicts the theory that defense can outpace offense on relevant timescales — essentially regardless of capability differences between the models.
- harrisedouard pushes back: if the two models' cyber capabilities differ enough, it isn't evidence about the theoretical argument; he adds that the biggest practical problem in cybersecurity is that too many vulnerabilities have no named owners anywhere in the economy.
More from AGI Musings
- Debugging the Human Mind Is the Ultimate AI Challenge, Researcher Quips — irinarish · 2026-10-23
- Security legend Halvar Flake: Rejecting AI-assisted bug reports over their origin is out of place — mboehme_ · 2026-10-23
- Distributed cognition exits stealth to build collective superintelligence — alexmacgregor__ · 2026-10-23
- DeepMind's Geoffrey Irving: alignment may need cryptography-grade rigor as AI finds 12 OpenSSL zero-days — geoffreyirving · 2026-10-23
- How WW2 made society worship intelligence — and why that era is ending — StewartalsopIII · 2026-10-23
- Oxford Academics Question Why Firms Race to Build Tech They Call a 'Doomsday Device' — SandraWachter5 · 2026-10-23