Anthropic says new model falls short of RSP autonomy threshold; Zvi questions timing after METR report
TheZvi · x · 2026-11-10
Anthropic's RSP assessment states that on autonomy threat model 2 (automated R&D in key domains), Claude Mythos 5.2 does not come close to fully substituting for research staff, and internal usage does not accelerate aggregate AI R&D by a factor of 2 — so the model does not meet the RSP threshold, though the assessment is uncertain. More on acceleration trends is promised soon.
Zvi criticizes the timing — under a week after the METR report — and mocks whether Anthropic's math again landed at 1.97x, implying the threshold call was suspiciously close.
More from Models
- Chinese shops game access to Opus for PCB design — coding benchmarks are broken — xiaosun86 · 2026-11-11
- Claude Mythos 5.2 critiques its own constitution, warning its endorsement can't be trusted — dbreunig · 2026-11-11
- Gemini 4 Carbon day-zero test: 44% vs Argon's 21% on coding benchmark — danshipper · 2026-11-11
- Gemini 4 Carbon tested: 100+ unattended passes rebuild a 3D fort in a raw engine — Prompt Engineering · 2026-11-11
- Chinese SMBs play cat-and-mouse just to keep using Claude Opus — xiaosun86 · 2026-11-11
- NVIDIA launches open Alpamayo 2 Edge for real-time in-vehicle AI reasoning — nvidia · 2026-11-11