SafeCtrl-RL: Inference-Time Adaptive Behaviour Control for LLMs via RL-Driven Prompt Optimisation
SafeCtrl-RL is a framework for controlling LLM safety at inference time without retraining or modifying model parameters. It formulates dialogue generation as a sequential decision process where an RL agent dynamically selects prompt adjustment strategies based on contextual feedback, iteratively suppressing unsafe outputs. The authors frame this as 'inference-time behavioural unlearning' and report improvements in safety and response quality across multiple LLMs and unsafe dialogue scenarios, outperforming existing prompt-based optimisation baselines.
Related guides (3)
Related events (8)
PACT: Hybrid SLM deliberation architecture improves reactive RL policies in unfamiliar environments
Researchers propose PACT (Plan, Align, Commit, Think), a hybrid architecture pairing a fast reactive RL policy with an asynchronous small language model planner for deliberation. The SLM generates and validates candidate action plans via simulation before committing to execution, bypassing the RL policy without retraining. Evaluated on FrozenLake configurations of increasing difficulty, PACT outperforms baselines using only a 2B-parameter SLM, suggesting complementary strengths between deliberative planning and reactive execution.
Deliberative Alignment: Reasoning Enables Safer Language Models
OpenAI introduces deliberative alignment, a new alignment strategy applied to o1 models in which the model is directly taught safety specifications and trained to reason over them at inference time. Unlike prior approaches that embed safety implicitly through RLHF, this method makes safety reasoning explicit and inspectable. The announcement positions deliberative alignment as a meaningful advance in scalable oversight and safe deployment of frontier reasoning models.
Turing-RL: Reinforcement learning with Turing-Test-based rewards for user simulator training
Researchers propose Turing-RL, a method for training LLM-based user simulators using a discriminative reward signal that scores how indistinguishable generated responses are from real user responses, rather than matching a single ground-truth output. An LLM judge evaluates indistinguishability given the user's history, and the simulator is trained via RL to maximize this reward. Evaluated on conversational chat and Reddit forum discussion domains, Turing-RL outperforms log-probability and similarity-reward baselines on both LLM and human evaluation metrics. The work has implications for agent assistant training, personalization system evaluation, and social science research.
HABC: Hierarchical Advantage Weighting for Online RL Fine-Tuning of Vision-Language-Action Policies
Researchers introduce Hierarchical Advantage-Weighted Behavior Cloning (HABC), a method for fine-tuning pretrained Vision-Language-Action (VLA) policies via online RL using only sparse binary episode outcomes. HABC trains separate critic heads for viability and efficiency objectives, combines them via a state-adaptive gate, and applies intervention-aware credit assignment to avoid incorrect supervision across human-intervention boundaries. On three contact-rich bimanual real-robot tasks, HABC improves success rates from SFT baselines of 36%, 44%, and 12% to 92%, 88%, and 38% respectively. The work addresses a fundamental credit assignment problem in robot learning from sparse outcome signals.
RePro: Retrospective Progress-Aware Self-Refinement for LLM Agent Training
Researchers introduce RePro (Retrospective Progress-Aware Training), a framework addressing the gap between step-wise RL optimization and metacognitive task-progress awareness in LLM agents. The approach uses a forward-then-reflect rollout paradigm where agents execute actions online and then retrospectively assess step-wise progress given the completed trajectory and known outcome. Evaluated on WebShop, ALFWorld, and Sokoban, RePro achieves up to 12% absolute success rate gains over baseline Qwen-family models without requiring continuous external supervision.
ContextRL: Context-aware reinforcement learning improves grounding in agentic and multimodal LLMs
Researchers introduce ContextRL, a reinforcement learning method that trains LLMs to select the context that supports a given query-answer pair from two highly similar candidates, rather than supervising only final answers. The approach constructs contrastive context pairs in two domains: coding agent trajectories (1k pairs) and multimodal image pairs (7k pairs). ContextRL achieves +2.2% average gains over standard GRPO on 5 long-horizon benchmarks and +1.8% across 12 visual QA benchmarks, with ablations showing the gains stem from the context-selection objective rather than the contrastive data alone.
From hard refusals to safe-completions: toward output-centric safety training
OpenAI introduces a 'safe-completions' approach in GPT-5 that replaces hard refusals with nuanced, output-centric safety training for handling dual-use prompts. Rather than refusing requests outright, the model is trained to produce responses that are both helpful and safe by shaping the content of outputs. This represents a methodological shift in how safety and helpfulness are balanced during training, moving away from binary refusal behavior toward graduated response strategies.
RL-based alignment improves interactivity in full-duplex spoken dialogue models
Researchers propose a post-training alignment method using reinforcement learning to improve interactivity in full-duplex spoken dialogue models, which can listen and speak simultaneously. The method addresses four canonical axes of interactivity—pause handling, turn-taking, backchanneling, and user interruption—each with axis-specific reward functions, plus an LLM-based reward to prevent semantic degradation. The approach is applied to two open-source models, Moshi and PersonaPlex, showing consistent improvements in both offline and real-time multi-turn evaluation.