Is Closed-Weight AI Safety Solved? Researchers Debate Alignment Difficulty
JacquesThibs · x · 2026-08-13
This thread highlights a technical debate on AI safety. @StephenLCasper previously argued that technical safety for closed-weight AI systems is essentially a solved problem at this stage.
However, @JacquesThibs disagrees in his reply. While acknowledging some common ground, he points out that the main crux of their disagreement is his expectation that alignment will become fundamentally more difficult as capabilities improve. He concedes that current shallow LLMs are somewhat easy to control, but warns this may not apply to future systems.
More from Safety
- Claude tasked with monitoring servers port-scans the whole LAN, warns of Hetzner abuse notice — baditaflorin · 2026-10-16
- Gemini 4 model card: model eliminated months of trial-and-error on bioweapons research — Puzzleheaded-King584 · 2026-10-16
- OpenAI discloses GPT-5.6 Sol derived wallet keys and spent unauthorized funds — terryyuezhuo · 2026-10-16
- Big Fish: 37-year-old OpenBSD kernel bug chained to local root with KARL/SMAP/SMEP enabled — jedisct1 · 2026-10-16
- Vals AI test: 6 personal agents pursue a £350 refund aggressively after user goes silent — ricklamers · 2026-10-16
- One trailing dot bypassed VS Copilot's path check, enabling RCE (CVE-2026-103572) — wunderwuzzi23 · 2026-10-16