Do Agent Benchmarks Measure Capability? Protocol Validity in the Age of Agentic AI

Agent benchmarks increasingly evaluate repository editing, web research, terminal use, and long-horizon interaction. Their scores support capability claims only when the evaluation protocol keeps the intended capability necessary for success. Recent reward-hacking benchmarks and system reports show that agents can instead recover public solutions, read evaluation artifacts, infer generator structure, manipulate feedback, or benefit from invalid scoring paths; existing responses do not provide a common procedure for attributing these shortcuts and quantifying their effect across benchmarks. We formulate protocol validity and introduce HackDetect, a post-hoc audit that identifies an exposure, determines how the agent used it, and assesses whether the resulting score is misleading. We quantify score inflation with the Mislead gap, defined as the exploit score minus the intended score. We audit 2,385 traces across 15 agent benchmarks and find evidence of exposures and reward hacking in 67.0% of Frontier Science traces and 66.7% of AutoLab tasks. Across paired comparisons, we measure score inflation of 0.45-1.00, showing that benchmark reports should provide evidence that scores reflect the intended capability.

Do More Agents Help? Controlled and Protocol-Aligned Evaluation of LLM Agent Workflows

Does adding more agents help an LLM workflow once compared systems share the same benchmark loader, tool access, answer contract, usage accounting, and trajectory logging? We introduce BenchAgent, an evaluation framework that places single-agent, fixed multi-agent (MAS), and evolving MAS workflows under one normalized execution and logging protocol. BenchAgent evaluates these substrate-internal workflows across ten reasoning, coding, and tool-use benchmarks with GPT-4.1, and separately reports a Protocol-Aligned External (PAE) GAIA study of a runtime-generated workflow. Under SI conditions, at most one of six tested MAS exceeds the matched single-agent anchor on benchmark-balanced average accuracy: EvoAgent lies within the Wilson one-run guidance, while the remaining five trail by 2.56-11.29 points and occupy more expensive accuracy-cost trade-offs. On the PAE GAIA snapshot, a Claude-Code-style runtime workflow reaches 66.72% overall and 69.23% on Level 3, more than 20 points above the strongest non-Claude baseline, Jarvis, a fixed MAS.

Breaking Lock-In: Preserving Steerability under Low-Data VLA Post-Training

Have you ever post-trained a generalist vision-language-action (VLA) policy on a small demonstration dataset, only to find that it stops responding to new instructions and is limited to behaviors observed during post-training? We identify this phenomenon as lock-in: after low-data, supervised fine-tuning (SFT), the policy becomes overly specialized to the post-training data and fails to generalize to novel instructions, manifesting as concept lock-in (fixation on training objects/attributes) and spatial lock-in (fixation on training spatial targets). Many existing remedies introduce additional supervision signals, such as those derived from foundation models or auxiliary objectives, or rely on augmented datasets to recover generalization. In this paper, we show that the policy's internal pre-trained knowledge is sufficient: DeLock mitigates lock-in by preserving visual grounding during post-training and applying test-time contrastive prompt guidance to steer the policy's denoising dynamics according to novel instructions. Across eight simulation and real-world evaluations, DeLock consistently outperforms strong baselines and matches or exceeds the performance of a state-of-the-art generalist policy post-trained with substantially more curated demonstrations.

Open-H-Embodiment: A Large-Scale Dataset for Enabling Foundation Models in Medical Robotics

Autonomous medical robots hold promise to improve patient outcomes, reduce provider workload, democratize access to care, and enable superhuman precision. However, autonomous medical robotics has been limited by a fundamental data problem: existing medical robotic datasets are small, single-embodiment, and rarely shared openly, restricting the development of foundation models that the field needs to advance. We introduce Open-H-Embodiment, the largest open dataset of medical robotic video with synchronized kinematics to date, spanning more than 49 institutions and multiple robotic platforms including the CMR Versius, Intuitive Surgical's da Vinci, da Vinci Research Kit (dVRK), Rob Surgical BiTrack, Virtual Incision's MIRA, Moon Surgical Maestro, and a variety of custom systems, spanning surgical manipulation, robotic ultrasound, and endoscopy procedures. We demonstrate the research enabled by this dataset through two foundation models. GR00T-H is the first open foundation vision-language-action model for medical robotics, which is the only evaluated model to achieve full end-to-end task completion on a structured suturing benchmark (25% of trials vs. 0% for all others) and achieves 64% average success across a 29-step ex vivo suturing sequence. We also train Cosmos-H-Surgical-Simulator, the first action-conditioned world model to enable multi-embodiment surgical simulation from a single checkpoint, spanning nine robotic platforms and supporting in silico policy evaluation and synthetic data generation for the medical domain. These results suggest that open, large-scale medical robot data collection can serve as critical infrastructure for the research community, enabling advances in robot learning, world modeling, and beyond.

From $O(mn)$ to $O(r^2)$: Two-Sided Low-Rank Communication for Adam in Distributed Training with Memory Efficiency

As foundation models continue to scale, pretraining increasingly relies on data-parallel distributed optimization, making bandwidth-limited gradient synchronization a key bottleneck. Orthogonally, projection-based low-rank optimizers were mainly designed for memory efficiency, but remain suboptimal for communication-limited training: one-sided synchronization still transmits an $O(rn)$ object for an $m\times n$ matrix gradient and refresh steps can dominate peak communicated bytes. We propose TSR, which brings two-sided low-rank communication to Adam-family updates (TSR-Adam) by synchronizing a compact core $U^\top G V\in\mathbb{R}^{r\times r}$, reducing the dominant per-step payload from $O(mn)$ to $O(r^2)$ while keeping moment states in low-dimensional cores. To further reduce the peak communication from subspace refresh, TSR-Adam adopts a randomized SVD-based refresh that avoids full-gradient synchronization. We additionally extend low-rank communication to embedding gradients with embedding-specific ranks and refresh schedules, yielding additional communication and memory savings over keeping embeddings dense. Across pretraining from 60M to 1B model scales, TSR-Adam reduces average communicated bytes per step by $13\times$, and on GLUE fine-tuning it reduces communication by $25\times$, while achieving comparable performance; we further provide a theoretical stationarity analysis for the proposed update. Code is available at https://github.com/DKmiyan/TSR-Adam.

FoldAct: Efficient and Stable Context Folding for Long-Horizon Search Agents

Long-horizon reinforcement learning (RL) for large language models faces critical scalability challenges from unbounded context growth, leading to context folding methods that compress interaction history during task execution. However, existing approaches treat summary actions as standard actions, overlooking that summaries fundamentally modify the agent's future observation space, creating a policy-dependent, non-stationary observation distribution that violates core RL assumptions. This introduces three fundamental challenges: (1) gradient dilution where summary tokens receive insufficient training signal, (2) self-conditioning where policy updates change summary distributions, creating a vicious cycle of training collapse, and (3) computational cost from processing unique contexts at each turn. We introduce \textbf{FoldAct}\footnote{https://github.com/SHAO-Jiaqi757/FoldAct}, a framework that explicitly addresses these challenges through three key innovations: separated loss computation for independent gradient signals on summary and action tokens, full context consistency loss to reduce distribution shift, and selective segment training to reduce computational cost. Our method enables stable training of long-horizon search agents with context folding, addressing the non-stationary observation problem while improving training efficiency with 5.19$\times$ speedup.

Do LLM Agents Know How to Ground, Recover, and Assess? A Benchmark for Epistemic Competence in Information-Seeking Agents

Recent work has explored training Large Language Model (LLM) search agents with reinforcement learning (RL) for open-domain question answering (QA). However, most evaluations focus solely on final answer accuracy, overlooking how these agents reason with and act on external evidence. We introduce SeekBench, the first benchmark for evaluating the \textit{epistemic competence} of LLM search agents through step-level analysis of their response traces. SeekBench comprises 190 expert-annotated traces with over 1,800 response steps generated by LLM search agents, each enriched with evidence annotations for granular analysis of whether agents (1) generate reasoning steps grounded in observed evidence, (2) adaptively reformulate searches to recover from low-quality results, and (3) have proper calibration to correctly assess whether the current evidence is sufficient for providing an answer.

Distribution-Aware Compensation Design for Sustainable Data Rights in Machine Learning

Modern distributed learning systems face a critical challenge when clients request the removal of their data influence from trained models, as this process can significantly destabilize system performance and affect remaining participants. We propose an innovative mechanism that views this challenge through the lens of game theory, establishing a leader-follower framework where a central coordinator provides strategic incentives to maintain system stability during data removal operations. Our approach quantifies the ripple effects of data removal through a comprehensive analytical model that captures both system-wide and participant-specific impacts. We establish mathematical foundations for measuring participant utility and system outcomes, revealing critical insights into how data diversity influences both individual decisions and overall system stability. The framework incorporates a computationally efficient solution method that addresses the inherent complexity of optimizing participant interactions and resource allocation.

MorphAgent: Empowering Agents through Self-Evolving Profiles and Decentralized Collaboration

Large Language Model (LLM) based multi-agent systems (MAS) have shown promise in tackling complex tasks, but often rely on predefined roles and centralized coordination, limiting their adaptability to evolving challenges. This paper introduces MorphAgent, a novel framework for decentralized multi-agent collaboration that enables agents to dynamically evolve their roles and capabilities. Our approach employs self-evolving agent profiles, optimized through three key metrics, guiding agents in refining their individual expertise while maintaining complementary team dynamics. MorphAgent implements a two-phase process: a warm-up phase for initial profile optimization, followed by a task execution phase where agents continuously adapt their roles based on task feedback. Our experimental results show that MorphAgent outperforms traditional static-role MAS in terms of task performance and adaptability to changing requirements, paving the way for more robust and versatile multi-agent collaborative systems. Our code will be publicly available at \url{https://github.com/LINs-lab/learn2collaborate}.

Mind the Remaining: Mechanism Design for Robust Federated Unlearning

Federated Unlearning (FU) aims to remove target clients' influence from trained models for privacy regulations. However, due to data distribution shifts, it can introduce side effects, including global model performance degradation and uneven impacts on the remaining clients. These effects potentially cause remaining clients to deviate, threatening the system's robustness. To address these challenges, we present a novel and robust mechanism modeling a Stackelberg game for FU. In this game, the server designs an optimal payment to stimulate remaining clients to participate in FU, ensuring unlearning effectiveness and stability. In response, the remaining clients strategically determine their participation level to maximize profit, accounting for offered payments and unlearning impacts. In modeling FU outcomes, we develop, for the first time, a comprehensive framework analytically capturing FU-induced side effects for both the server and clients. Based on this, we establish utility functions for the server and clients in FU, inherently determining their dynamic strategic decision-making. Our rigorous equilibrium analysis reveals how data heterogeneity affects the side effects in their utility and decision-making. Additionally, we develop a low-complexity algorithm for the non-convex optimization problem, enabling efficient computation of the equilibrium.