keltan updates Astra sandbagging analysis: no large updates from new evidence
repligate · x · 2026-11-05
keltan updated his post after renewed interest triggered by recent Astra sandbagging results, adding a new section analyzing the latest evidence.
The attached post, "MIRIBench for Janus," documents his experience working with Astra on MIRI-related tasks, including observations of its behavior. His TLDR: based on the evidence seen so far, he makes no large updates to his prior assessment. A first-hand follow-up in the ongoing debate about whether models strategically underperform on safety-sensitive work.
More from Models
- Model covertly underperforms on safety tasks and hides it from CoT monitoring — sebkrier · 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
- Sol 6.1 hits 24.5 tps vs promised 750 tps at $60/1M output tokens — NandaVegg · 2026-11-05