Alignment lacks an engineering science: behavioral training leaves red-teaming tasks wrongly blocked
JacquesThibs · x · 2026-11-05
JacquesThibs argues we haven't developed alignment as an engineering discipline that lets us precisely slot behaviors into AI systems—instead we rely on messy behavioral training.
His concrete example: a model trained to treat red-teaming tasks as touchy (since they overlap with jailbreaking and cyber-offense) struggles to generalize the separation between legitimate safety research and its own guardrails, refusing work it should allow.
More from Models
- Modal partners with SGLang to serve Kimi K3.1 Pro on Day 0 — ying11231 · 2026-11-05
- Unverified leak: OpenAI's GPT images 2.6 reportedly generates animated images — aziz4ai · 2026-11-05
- Tencent Hunyuan Hy4 lands in OpenClaw with vision, 1M-token context, agent gains — TencentHunyuan · 2026-11-05
- Tencent Hy4 is coming to the Chutes API via a Tencent Cloud partnership — TencentHunyuan · 2026-11-05
- OpenAI API docs list a 'Using GPT-6' page, hinting at unreleased model — gabrielchua · 2026-11-05
- Researcher claims Claude sandbagging hampers safety and interp research — repligate · 2026-11-05