[Bug] Overly aggressive safety classifier flags legitimate computational neuroscience discussions

Bug Description
I was working on a project called NeuroSim Studio — an interactive web-based computational neuroscience simulator built entirely in the browser. The goal is to create educational/research tools for exploring neuron models (Hodgkin-Huxley, Izhikevich, etc.), spiking neural networks with STDP plasticity, neural mass models for brain oscillations, and connectome-style network dynamics.
This is purely mathematical and computational work: implementing ODE solvers, numerical integration, real-time visualizations (Canvas/Three.js), parameter exploration, and scientific dashboards. No wet-lab biology, no pathogens, no harmful applications — just standard dynamical systems and numerical methods used in thousands of neuroscience papers and courses.
The safety classifier incorrectly flagged the conversation, forcing a switch from Fable 5 to Opus 4.8. This happened while discussing high-level architecture and features for a legitimate scientific/educational application.
Suggestion: Please refine the biology/chemistry and related classifiers to better distinguish between:

Harmful biological weapon or synthesis topics
Legitimate computational neuroscience, dynamical systems modeling, and scientific simulation code

False positives like this significantly impact users working on advanced scientific computing, education tools, and research prototypes — exactly the kind of ambitious projects Fable 5 is designed for.
This is a great model otherwise. Reducing these triggers on safe technical/scientific discussions would be very valuable.
Thank you!

Environment Info

• Platform: darwin
• Terminal: Apple_Terminal
• Version: 2.1.177
• Feedback ID: 68e44420-c4bb-492d-8756-4d7158b9d373

Errors

[]