TensorCommitments proposes a tensor-native proof-of-inference scheme for verifiable LLM inference, reducing the need for trust in remote GPU execution and improving robustness to LLM attacks.

This paper explores a novel method of inducing vulnerabilities in large language models (LLMs) using 'drunk language', which can lead to jailbreaking and privacy leaks. The researchers found that LLMs are more susceptible to these vulnerabilities than previously reported approaches.

VISPA is a training-free framework for pluralistic alignment of large language models, enabling direct control over value expression by dynamic selection and internal model activation steering. It achieves performant results across various pluralistic alignment modes in healthcare and beyond, and is adaptable with different steering initiations, models, and/or values.

Researchers evaluated 7 TDD methods on 8 CodeLLMs, introducing CodeSnitch, a function-level benchmark dataset. The study assessed robustness under code clone detection taxonomy.

Researchers propose UniGuardian, a training-free detector for Large Language Models (LLMs) that identifies prompt injection, backdoor, and adversarial attacks without knowing the attack type. UniGuardian measures how prompt perturbations shift the model's output distribution and uses a single-forward strategy for efficient detection and text generation.

Researchers propose POEF, an automated red-teaming framework to bridge the intent-behavior gap in LLM-based robot jailbreaks, demonstrating an 80% behavior jailbreak success rate.

A new approach, Loki, adapts pretrained models to predict new classes without additional training by using a metric that relates labels via distances. This can improve model performance on zero-shot prediction tasks.

ESCROW is a post-deployment maintenance framework for LLM agents in policy-governed enterprise workflows. It updates an agent's external skills under a strict update boundary, ensuring reliable and auditable changes.

A new framework called Semantic Cooperative Games (SCG) is proposed for contribution attribution in LLM-based multi-agent systems. SCG represents a language flow as a semantic generation hypergraph and computes an agent-level semantic value function. It introduces a new method called SLIC to allocate contributions without rerunning agent subsets, reducing computation cost by 93.3%.

This paper presents a predictive law for calculating the uplift of Large Language Model (LLM) ensemble performance based on diversity of thought. It provides an experimentally verified formal law and a compact heuristic for calculating uplift, which is tested on various datasets.

MINCE is a method for shrinking LLM evaluation datasets by using Monte Carlo simulation to find the minimum subset size that bounds accuracy drift, reducing evaluation time by up to 89%.

Researchers investigated the performance of embodied LLMs in a physical robotic setup with varying levels of observation fidelity. They found that LLMs performed best under raw RGB input and worst under perfect ground-truth observations. This suggests that measured performance may not reflect robust problem-solving abilities, but rather the interaction between perceptual errors and reasoning failures.

A new adaptive skill-reuse framework, SkillLens, is proposed for cost-efficient LLM agents. It organizes skills into a hierarchical graph and retrieves them at mixed granularity to improve task performance and reduce costs.

A dataset of real-world coding agent sessions from open-source developers is presented, providing an empirical characterization of agent usage and failure modes. The dataset shows that agents remain inefficient in natural settings and introduce more security vulnerabilities than human-authored code.

Researchers propose a benchmark (NARCBench) and probing techniques for detecting collusion between AI agents in multi-agent systems, achieving high detection rates in various scenarios.

Researchers propose Pheromone-Guided Policy Optimization (PhGPO) to improve long-horizon tool planning for Large Language Model (LLM) agents, leveraging historical trajectories to guide policy optimization and improve tool transitions.

Researchers propose Scalable Delphi, a method for using large language models to estimate structured risk by adapting the Delphi method for LLMs with diverse expert personas, iterative refinement, and rationale sharing.

Researchers propose Graph of Concept Predictors (GCP), a reasoning-aware active distillation framework for Large Language Models (LLMs) that improves performance under limited annotation budgets while yielding more interpretable and controllable training dynamics.

A benchmarking framework, AstroAgentBench, is introduced for evaluating agentic planning in space mission planning tasks. It assesses the performance of large language model (LLM) agents in domains like scheduling, observation planning, and constellation design. The framework provides a standardized evaluation methodology and highlights the importance of task-contract formulation, verifier feedback, and search adaptation in achieving high-quality plans.

KaliBench is a fine-grained benchmark for evaluating the ability of language models to generate executable commands for real-world cybersecurity tools. It includes 8,504 query-command pairs across 1,642 tools and enables precise and reproducible assessment of tool selection and argument construction.