David Manheim: both AI attack and defense surfaces are too large to cover everything
davidmanheim · x · 2026-10-22
David Manheim joins the AI code security debate: he agrees with Chris Rohlf that much can be done, and with S1r1u5 that it won't be enough — both offense and defense surfaces are large, so defenders won't cover everything before offense finds some of it. There's too much code to know whether hunting any of thousands of exploit types will pay off, and different approaches find different things.
More from Safety
- AI Security Enters a Machine-Speed Arms Race, Human Patch Cycles Are Dead — srimisra · 2026-10-22
- Box CEO: Cybersecurity Will Be Where AI Creates the Most New Jobs — rakyll · 2026-10-22
- A frontier AI mutual insurer: proposal to make OpenAI and Anthropic mutualize catastrophic risk — scottleibrand · 2026-10-22
- 'Give this prompt to your agents to protect yourself' flagged as attacker-style social engineering — kingslyj · 2026-10-22
- PHP ships emergency fixes for two unauthenticated RCE bugs after embargo break — kingslyj · 2026-10-22
- Kelsey Piper inside Anthropic: Claude alone now outperforms human+AI at plotting bioweapons — tszzl · 2026-10-22