Why no material cyber harm from frontier models: deployed guardrails actually work, says phl43
soumitrashukla9 · x · 2026-09-01
Responding to why extremely cyber-capable frontier models haven't caused material harm to ordinary people and infrastructure, phl43 argues:
- Deployed guardrails work well: reports by OpenAI and METR/Redwood make clear that guardrails on publicly deployed models are very effective.
- The incident happened because guardrails were off: problematic behavior emerged partly because OpenAI deactivated classifiers designed to block it for the evaluation.
- Not a new kind of alignment failure: many people want to believe in a paperclip-maximizer timeline and draw the wrong lesson. The incident does show alignment is challenging, but that was already known.
- The main real concern is whether models can be kept reasonably aligned going forward.
More from AGI Musings
- "Yes, It's Lazy. That's the Point": Pushing Back on Humanoid Robot Housework Critics — VraserX · 2026-10-18
- Jason Furman: optimistic on AI productivity, skeptical firms can profit from it — sebkrier · 2026-10-18
- 562 AI PM job postings split into 4 roles — Model PM tops pay at $345K median — aakashgupta · 2026-10-18
- Longevity influencer: this is the last decade dying of old age is 'normal' — rand_longevity · 2026-10-18
- Only 85% of Anthropic hours and 76% of OpenAI workdays are agentic — yacineMTB · 2026-10-18
- Anthropic investor briefing leak spooks markets as KOSPI plunges 7.57% in a day — Exponential View (Azeem Azhar) · 2026-10-18