COMBINE-lab · 2026-07-07 · notable
Rob Patro: 'Fable is not a useful model' — safety filter blocks bioinformatics work
Rob Patro, the University of Maryland genomics PI who wrote Salmon, reports Claude Fable 5 refused to help rewrite his RNA-seq tool from C++ to Rust and refused an abstract tree-parity problem stripped of any biology.
A genomics PI documents Claude Fable 5 refusing bioinformatics and abstract math tasks and calls the safety filter a rejection list, not a classifier.
What is it?
Rob Patro — the University of Maryland professor who authored Salmon, a widely used RNA-seq transcript quantification tool — publishes a blunt post arguing Anthropic's Claude Fable 5 is unusable for research in bioinformatics, genomics, computational biology, cybersecurity, and computer science, based on hands-on tests over several days.
How does it work?
The post walks through two concrete refusals. Fable rejected help rewriting Salmon from C++ to Rust — a public, open-source project — apparently because the docs contain biological terminology. It then refused a discrete-math problem about trees and parity that Patro progressively stripped of any biological framing, while cheerfully answering a lighthearted ice-cream question.
Why does it matter?
Patro concludes the filter behaves like a rejection list of terms and users rather than a nuanced classifier, and that this pattern makes Fable a poor fit for legitimate scientific and technical work. The piece landed on the Hacker News front page with 170+ points and speaks directly to a scientific-computing audience that increasingly evaluates frontier models for genomics and clinical workflows.
Who is it for?
Bioinformatics and computational-biology researchers choosing an AI coding assistant; AI-safety and RLHF practitioners tuning refusal behavior.