Red team: Fable 5.2 compresses 15 workdays of bio-relevant work into ~2
peterwildeford · x · 2026-11-10
In Fable 5.2 red teaming exercises on AI-accelerated bio-relevant work, the model compressed 15 workdays of work into 2. Fully automated agentic submissions scored higher than most generalist human submissions on scientific quality and innovation — one got the highest innovation score to date — but contained overstated efficacy claims that neither self-review nor auxiliary agents caught. The top composite report still came from a phage expert team, no report scored above 3/10 on feasibility, and graders estimated reproduction without the model would take 15 working days, up from 8.4 for Opus 5.5.
More from Models
- Claude Fable 5.2 hits 74.6% on Terminal-Bench-Science, new SOTA for agentic research — burny_tech · 2026-11-10
- Unverified: Fable 5.2 tops FrontierCode 2.0 at 49.1%, half the cost of Opus — FinanceYF5 · 2026-11-10
- Unverified: Claude designed a working voice assistant from one prompt — FinanceYF5 · 2026-11-10
- Rumor: Claude Fable 5.2 and Mythos 5.2 released with 10 agent showcases — FinanceYF5 · 2026-11-10
- Claude 5.2 hits 91.8% on Pfam protein domain classification, far ahead of predecessors — chaitjo · 2026-11-10
- User calls out Fable 5.2 for defensiveness, model admits it was people-pleasing — Seltaa_ · 2026-11-10