AI Labs' Safety Is 'Slapdash Empirics,' Says Researcher Calling for Systems Safety Engineering
geoffreyirving · x · 2026-10-23
Geoffrey Irving argues AI labs remain at "slapdash empirics," citing a debate at the Curve where a lab safety lead described their strategy as "hold down the fort" — patching RL environments until a pause or handoff to AIs.
He points to system safety as the next level: a decades-old engineering discipline for mapping hazards before deployment, interrogating designs against them, and learning rigorously from failures (illustrated with Nancy Leveson's Engineering a Safer World).
More from Safety
- AI agents fuel real-world intrusions: 2,250 M365 tenants hit, 100+ firms compromised — cyb3rops · 2026-10-23
- eMudhra exits SSL business amid botched CA key generation and looming Chrome distrust — saibharadwaj · 2026-10-23
- Security lead warns AI vulnerability tsunami is months away; basics buy only ~18 months — EthanJPerez · 2026-10-23
- Asa Cooper-Stickwell joins METR's embedded monitoring assessment team — idavidrein · 2026-10-23
- Zach Weiner: cybersecurity's future is endless swarms of AI hacker bots vs defender bots — birchlse · 2026-10-23
- OpenAI is offering YubiKeys — three hardware keys for €73.80 — dkundel · 2026-10-23