Contrastive Reasoning Alignment: Reinforcement Learning from Hidden Representations

📰 ArXiv cs.AI

Learn to improve robustness of large reasoning models against jailbreak attacks using Contrastive Reasoning Alignment (CRAFT) and reinforcement learning from hidden representations

advanced Published 20 May 2026
Action Steps
  1. Implement CRAFT, a red-teaming alignment framework, to align large reasoning models with safety-aware reasoning traces
  2. Define objectives over the hidden state space to optimize model robustness
  3. Use contrastive representation learning to improve model representations
  4. Apply reinforcement learning to update model parameters and improve alignment
  5. Evaluate model performance using metrics such as robustness and safety awareness
Who Needs to Know This

AI researchers and engineers working on reinforcement learning, robustness, and alignment of large language models can benefit from this technique to improve model safety and security

Key Insight

💡 Contrastive Reasoning Alignment (CRAFT) can improve robustness of large reasoning models by optimizing objectives over the hidden state space

Share This
🚀 Improve robustness of large reasoning models against jailbreak attacks with CRAFT, a novel alignment framework #AI #ReinforcementLearning #Robustness

Key Takeaways

Learn to improve robustness of large reasoning models against jailbreak attacks using Contrastive Reasoning Alignment (CRAFT) and reinforcement learning from hidden representations

Full Article

Title: Contrastive Reasoning Alignment: Reinforcement Learning from Hidden Representations

Abstract:
arXiv:2603.17305v2 Announce Type: replace Abstract: We propose CRAFT, a red-teaming alignment framework that leverages model reasoning capabilities and hidden representations to improve robustness against jailbreak attacks. Unlike prior defenses that operate primarily at the output level, CRAFT aligns large reasoning models to generate safety-aware reasoning traces by explicitly optimizing objectives defined over the hidden state space. Methodologically, CRAFT integrates contrastive representati
Read full paper → ← Back to Reads

Related Videos

Build AI Agents in 2 Minutes using Microsoft Foundry
Build AI Agents in 2 Minutes using Microsoft Foundry
Rajeev Kanth | BEPEC
Evaluating Agentic AI Skills (using OpenHands)
Evaluating Agentic AI Skills (using OpenHands)
Rajistics - data science, AI, and machine learning
Dynamic Workflows using Openhands SDK
Dynamic Workflows using Openhands SDK
Rajistics - data science, AI, and machine learning
I built a custom Hermes plugin. #HermesAgent #Claudecode #openaicodex #openclaw #nousresearch
I built a custom Hermes plugin. #HermesAgent #Claudecode #openaicodex #openclaw #nousresearch
Tech Friend AJ
I Tried Hermes Desktop. It Might Replace My AI Agent Setup
I Tried Hermes Desktop. It Might Replace My AI Agent Setup
Tech Friend AJ
Trying Hermes Desktop + More
Trying Hermes Desktop + More
Tech Friend AJ