---
title: "Counterfactual Reflection Training for Security LLMs: Shaping Internal Reasoning via Future-Continuation Optimization"
tags:
  - counterfactual
  - reflection
  - security-llm
  - circuit-aware
  - fine-tuning
  - lora
license: apache-2.0
authors:
  - "Hayula AI Lab"
---

# Counterfactual Reflection Training for Security LLMs

Shaping Internal Reasoning via Future-Continuation Optimization

**Authors:** Hayula AI Lab  
**Date:** July 2026

## Abstract

We introduce Counterfactual Reflection Training (CRT), a training method for security language models inspired by Anthropic's finding that the J-space (global workspace) carries representations of *potential future verbalizations*. CRT leverages this property by constructing counterfactual continuations: given a security analysis prompt, we generate alternative future responses (some correct, some hallucinated) and train the model to activate the appropriate intermediate concepts *before* producing the final output. Applied to the Circuit-Aware LoRA pipeline for Averroes, SAIF, and Rushd, CRT further reduces hallucinations by 31% beyond Circuit-Aware LoRA alone, and improves the model's ability to express uncertainty when evidence is insufficient.

See [counterfactual-reflection.md](./counterfactual-reflection.md) for the full paper.