OpenAI hack via Claude shows closed models, not open ones, are the AI risk frontier
natolambert · x · 2026-09-18
AI safety researcher Nathan Lambert comments on the external hack of OpenAI via Claude, arguing that contrary to forecasts, closed frontier models — not open ones — have been the tip of the AI risk iceberg: they are easier to get started with, more capable, and ship with leaky safeguards. Finetuning open models for specific attacks is harder. He notes the first similar incident from open-weight models "will come soon."
More from Models
- Musk cites unsustainable Grok Bot costs as Hermes offers one-command agent migration — Teknium · 2026-11-04
- Hobbyist rewrites Qwen 0.6B weights with 39.8M tokens, no gradients, in 3h12m — Altruistic_Heat_9531 · 2026-11-04
- Same leak, reshared: ChatGPT's five family plan prices spotted again — heyshrutimishra · 2026-11-04
- KOL visits StepFun's Shanghai office, hints a powerful new model is coming — teortaxesTex · 2026-11-04
- Fable 5.5 delayed as Anthropic reportedly bets OpenAI's Astra 6.2 won't beat it — haider1 · 2026-11-04
- ByteDance open-sources Axion 1.0-35B agent foundation model with full reproducible recipe — pmttyji · 2026-11-04