arXiv:2606.15491v1 Announce Type: new Abstract: Radar SLAM is attractive for autonomous ground vehicles operating in visually degraded environments, however, scanning radars are noisy, have low scanning rates, and their measurements are challenging to match reliably over long trajectories. This paper presents FD-SLAM, a fast dense radar-inertial SLAM system that extends dense radar-inertial odometry with frequency-domain loop closure and pose graph optimization. The proposed method preserves an image-like structure of scanning radar measurements by using a compact frequency-domain polar descriptor for loop-candidate retrieval and a multi-stage verification pipeline based on temporal filtering, phase-correlation screening, scan-alignment similarity, and geometric consistency checks. Verified loop closures are added as non-sequential constraints in an SE(2) pose graph together with radar-inertial odometry factors. FD-SLAM is evaluated on a publicly available dataset using standard KITTI evaluation metrics. The results show that FD-SLAM improves FD-RIO baseline, achieves competitive performance against current state-of-the-art radar SLAM methods, and provides favorable rotational accuracy across multiple evaluated driving trajectories. Runtime analysis further indicates that the radar-inertial front-end operates above the radar sampling rate on a CPU-only setup, while loop closure detection and graph optimization remain suitable for parallel background execution.
Science Journals
arXiv:2606.15493v1 Announce Type: new Abstract: Model stealing attacks, where adversaries create high-fidelity surrogate models, are a significant threat to the intellectual property of machine learning services. Conventional wisdom suggests these surrogates could provide adversaries with economic leverage comparable to the original service providers. This paper challenges this assumption by evaluating model stealing attacks beyond mere fidelity to the target model. Because query-based extraction provides only partial supervision of the target's input-output behavior, the surrogate is not uniquely identified: many near-optimal surrogates can achieve comparable fidelity while differing in deployment-relevant properties. Instead of performing a classic learning-based model stealing attack, we compute the Rashomon Set (i.e., the set of almost-equally-accurate models) of surrogate models, and evaluate its diversity using multiplicity metrics (ambiguity, discrepancy, and Rashomon Capacity) and group fairness metrics. Across tabular, medical imaging, and NLP tasks, our experiments on real-world datasets reveal that despite exhibiting similar fidelity to the target model, surrogate models can display significant variances in other critical performance metrics. These findings cast doubt on the presumed equivalence between high-fidelity surrogates and the target model in practical deployment scenarios.
arXiv:2606.15494v1 Announce Type: new Abstract: Human-robot collaboration (HRC) in prefabricated construction requires planning approaches that consider not only productivity but also time-dependent worker states during repeated work and rest. Existing planning models often rely on simplified assumptions about fatigue, workload, or recovery, with limited domain-specific empirical evidence on how perceived strain evolves. This study develops an empirically grounded, planning-oriented approach to characterize perceived strain accumulation and recovery in prefabricated construction HRC. A controlled repeated work-rest experiment assessed perceived cognitive and physical strain using the Rating Scale for Mental Effort and Borg's Rating of Perceived Exertion. Linear and exponential functional forms were evaluated, followed by mixed-effects modeling to examine collaborative conditions, session effects, and inter-individual variability. Results indicate that cognitive strain accumulation is best represented by a linear mixed-effects model, whereas rest-phase recovery follows nonlinear decay. The resulting planning-oriented models may inform future human-state-aware task allocation and scheduling research.
arXiv:2606.15495v1 Announce Type: new Abstract: Identifying collective variables that capture slow dynamical modes is essential for sampling rare events in complex systems. Existing machine-learning approaches often require predefined metastable states, carefully chosen descriptors, or training trajectories with high-quality kinetic information. Here, we introduce SelfTICA, a self-supervised contrastive-learning framework that reformulates collective-variable discovery as dynamical representation learning. SelfTICA defines positive and negative pairs from time-lagged molecular configurations, learns reusable features through a contrastive objective linked to spectral variational principles, and extracts orthogonal slow modes by applying time-lagged independent component analysis in the learned representation space. By decoupling representation learning from slow-mode extraction, SelfTICA avoids direct optimization of eigendecomposition-based objectives and enables spectra and collective variables to be evaluated across lag times without retraining. Across different atomistic systems, SelfTICA learns dynamical representations from limited, biased, or exploratory data and converts them into collective variables that accelerate rare-event exploration and improve free-energy convergence.
arXiv:2606.15737v1 Announce Type: new Abstract: We overview applications of Craig interpolation and Beth definability to simplifying logical expressions or database queries. From the perspective of the theory of interpolation and definability the results give a number of new angles. First, they give a different take on what it means to make definability or interpolation results effective, looking at algorithms that take a proof as input and return an interpolant or explicit definition as output. Secondly, they relate interpolation and definability to preservation theorems in model theory: interpolation and definability theorems are the basis for many "semantics-to-syntax" results, relating a semantic property of a formula to its equivalence with a certain syntactic form. Thirdly, they motivate new forms of interpolation and definability, focusing on syntactic forms that are of interest in databases.
arXiv:2606.15497v1 Announce Type: new Abstract: The automation of science is a long-standing ambition in the field of AI. While the community has made significant progress in automating individual components of the scientific process, a system that autonomously navigates the entire research lifecycle -- from conception to publication -- has remained out of reach. Here, we present the strongest demonstration to date toward automating the entire process end-to-end. We present The AI Scientist, which creates research ideas, writes code, runs experiments, plots and analyzes data, writes the entire scientific manuscript and performs its own peer review. Its ideas, execution, and presentation are of sufficient quality to produce a manuscript generated by an AI system that passes the first round of peer review at a major machine learning conference workshop. The workshop has an acceptance rate of 70 percent. Our system leverages modern foundation models within a complex agentic system. We evaluate The AI Scientist in two settings: a focused mode using human-provided code templates as an initial scaffold to conduct research on a specific topic, and a template-free, open-ended mode that leverages agentic search for wider scientific exploration. Both settings produce diverse ideas and automatically test, report on, and evaluate them. This achievement demonstrates AI's growing capacity for scientific contribution and signifies a potential paradigm shift in how research is conducted. As with any impactful new technology, there could be significant risks, including taxing overwhelmed review systems and adding noise to scientific literature. However, if developed responsibly, such autonomous systems could greatly accelerate scientific discovery.
arXiv:2606.15499v1 Announce Type: new Abstract: Assembly theory is an experimental and theoretical framework that introduces a metrological approach to detecting life, with potential applications across diverse substrates. Its two central observables are assembly index and copy number. The assembly index is the minimum number of joining operations required to construct an object from its elementary parts; for molecules, it can be measured using mass spectrometry, infrared spectroscopy, and NMR. Copy number is the abundance of a given distinguishable object within a sample. A key empirical result of the theory is that high assembly index combined with high copy number constitutes a signature that cannot arise abiotically, and this has been validated experimentally in application to molecular biosignatures. The foundational theoretical concept underlying these results is the assembly space, which encodes the causal possibilities determinable from observed objects, with the assembly index the shortest path to them given the physical constraints of a given substrate. Here, we provide a generalized formalism to describe assembly spaces and tools for assembly index approximations. We begin by reviewing the applications of assembly theory across molecules, minerals and atmospheres, and then introduce a general, substrate-independent formal definition of assembly spaces and assembly indices. We develop a unified path hierarchy framework to clarify relationships among the various representations of assembly spaces and assembly paths that appear in the literature on molecular assembly. Finally, we show how formal grammar algorithms can be adapted to efficiently bound assembly index calculations and provide clarification on the utility of such approximations, with the goal to increase the accessibility of tools to explore this emerging area for a broader group of researchers across chemistry, biology, and complexity science.
arXiv:2606.15746v1 Announce Type: new Abstract: Timely information is important in a wide variety of Internet of Things (IoT) services in which a shared server must manage two competing tasks: (i) processing a queue of jobs, and (ii) generating status updates to a remote monitor. This creates a fundamental trade-off between queue stability and data freshness. In this work, we model this scheduling decision as a Markov Decision Process (MDP) with the objective of minimizing a weighted sum of the average Age of Information (AoI) and the average queue length. We show that the optimal scheduling strategy is a queue-dependent age threshold which is monotonic. The shape of the switching curve differs according to different priority regimes. Finally, we compare the optimal MDP policy against heuristic policies.
arXiv:2606.15505v1 Announce Type: new Abstract: Discovering the governing equations of a physical system from data is a central goal across the sciences, yet in most experiments only a few states are accessible while the rest stay hidden. Existing approaches treat this partial observability as an obstacle to be removed by first reconstructing the hidden state -- a step that is ill-posed under noise and that discards the physical constraints, such as energy conservation, that the true dynamics obey. We show that for conservative (Hamiltonian) systems no reconstruction is needed: projecting the dynamics onto the measured coordinates yields a memory kernel that we prove to be a lossless positive-real rational matrix, whose poles are the hidden natural frequencies and whose positive-semidefinite residues encode the couplings. The governing equation -- and the underlying Hamiltonian -- can therefore be read directly from the autocorrelation of the measured signal, with guarantees of uniqueness and physical passivity, and without neural networks. We validate the approach on linear, nonlinear, and chaotic systems under realistic noise. By recovering interpretable equations of motion that conserve energy by construction from partial measurements, the method offers a common tool for problems spanning mechanics, fluid and plasma physics, and beyond.
arXiv:2606.15508v1 Announce Type: new Abstract: Tool-augmented large language model agents increasingly operate over large tool libraries, but existing evaluations often focus on whether a model can call a tool correctly rather than how the visible tool menu shapes reliability, efficiency, and safety-relevant risk exposure. We introduce ToolMenuBench, a benchmark for evaluating tool-menu construction in multi-step LLM agents. ToolMenuBench varies tool-menu size, distractor type, state-dependent task structure, and risk exposure, and reports both filter-level and downstream agent metrics, including visible-tool count, risky-tool exposure, task success, wrong-tool calls, premature actions, and token usage. In a controlled evaluation across seven model backends, three tool-menu sizes, six filtering methods, and seven evaluation settings, CMTF improves task success from 32.1% under all-tools exposure to 85.7%, while reducing average token usage by roughly 98%. Causal minimal tool filtering achieves the strongest overall tradeoff, reducing visible tools, wrong-tool calls, premature actions, and risky-tool exposure relative to unfiltered exposure, lexical filtering, state-aware filtering, and broader causal-path baselines. ToolMenuBench provides a reusable evaluation framework for studying the agent-interface problem: which tools should be visible, when they should be visible, and under what cost or risk constraints.
arXiv:2606.15509v1 Announce Type: new Abstract: The concept of 'collaboration' has been extended rapidly to describe what people now do with conversational agents, intelligent tutors, adaptive platforms, and generative artificial intelligence (AI) tools in general. This chapter asks what is gained and lost when a demanding concept from the learning sciences is applied so freely. Returning to long-standing accounts of collaborative learning, it reconstructs the requirements that a situation, an interaction, and a set of cognitive processes have historically had to meet before being called collaborative. Human-AI collaboration requires a partly symmetric and negotiated relationship, shared and negotiable goals, a low and shifting division of labour, interactive and synchronous exchange, and mutual modelling, grounding, and socially shared regulation. Reviewing process-sensitive empirical studies of writing and problem solving, the chapter shows that most current human-AI interaction is better described as consultation, governance, delegation, or instruction rather than as collaboration. To make these distinctions functional, the chapter introduces a five-level diagnostic taxonomy of human-AI teaming (i.e. transactional, situational, operational, praxical, and synergistic) defined by the affordances an AI system exhibits. It shows that only the highest level begins to satisfy the conditions the tradition places on collaboration. The chapter derives the functions an AI system must possess for collaboration to be achievable, argues that most of these are present-day engineering choices rather than capabilities to be awaited, and sets out the implications for research, measurement, and responsible practice of human-AI collaboration in education.
arXiv:2606.15650v1 Announce Type: new Abstract: We present AnonShield, a high-throughput, on-premise pseudonymization system that combines GPU-accelerated NER, streaming processing, caching, and schema-aware configuration. Evaluated on datasets up to 550 MB (70,951 records), AnonShield reduces processing time from over 92 hours to under 10 minutes (up to 738x speedup) while achieving up to 94.2% F1-score and 96.7% recall. Our results show that scalable pseudonymization of vulnerability data is feasible without sacrificing analytical utility, enabling compliant data sharing in operational CSIRT environments.
arXiv:2606.15521v1 Announce Type: new Abstract: Tokenization introduces representational redundancy: under a fixed token vocabulary, every byte string admits many valid token encodings, or segmentations, that decode to the same surface string. However, given a prompt, most language model tokenizers break this representational symmetry by returning a canonical segmentation. Training only on canonical segmentations should influence inference behavior, and there is little reason to expect models to respect segmentation symmetry on downstream tasks. We find that this symmetry partially emerges during training. Here, we probe this emergent symmetry through experiments testing token compositional understanding, representation diversity, and task focused benchmark performance. We primarily use \textbf{retokenization} -- replacing a prompt's canonical tokenization with an alternative segmentation while preserving its bytes exactly. Relative to other prompt perturbations, retokenization is unusually clean because it isolates segmentation effects without changing syntax, semantics or surface form. We use retokenization to study sensitivity and robustness to semantically identical input representations across pretraining and post-training. Moreover, this partial retokenization symmetry suggests a distinct inference-time sampling axis. While temperature sampling generates diverse outputs from the model using its next-token probability distribution, retokenization generates diversity from the model's internal computations through semantically equivalent input representations. We find that while this retokenization sampling strategy can hurt performance on easy problems, it can also recover solutions that conventional sampling does not find. Overall, our work presents retokenization as a simple yet powerful probe of large language models, shedding light on compositional understanding and prompt sensitivity, and offering a novel sampling strategy.
arXiv:2606.15523v1 Announce Type: new Abstract: Spiking Vision Transformers (SViTs) have emerged as alternative low-power ViT models, but their large sizes hinder their deployments on resource-constrained embedded AI systems. To address this, state-of-the-art works proposed quantization techniques to compress SViT models, but their manual, human-guided approach needs a huge design time and power/energy consumption to find the appropriate quantization setting for each given network, making this approach not scalable for quantizing multiple networks. Toward this, we propose AQ4SViT, a novel automated quantization framework for SViTs that can provide quick quantization settings with good trade-offs between accuracy and memory. To achieve this, AQ4SViT employs the following key ideas: quantization search strategy that evaluates the quantization setting candidates while considering the accuracy constraint; and search gating policy that quickly evaluates and selects promising quantization candidates by leveraging membrane potential drift as a performance proxy. In the search gating policy, AQSViT employs two search algorithm variants to provide trade-off options: Greedy search, which performs fast but may lead to local optima; and Beam search, which performs slower but has better performance in finding global optima selection due to a wider search space. Experimental results show that AQ4SViT-Greedy quickly finds the appropriate quantization settings, achieving up to 6.6x faster search time and up to 82.5% memory saving compared to the state-of-the-art; while AQ4SViT-Beam further reduces the memory footprint by up to 90% compared to the state-of-the-art, but with 4.5x longer search time; all these results are obtained while maintaining high accuracy within 1.5% from the original/non-quantized models on the ImageNet dataset. These results highlight that AQ4SViT framework offers advancements toward SViT deployments on embedded AI systems.
arXiv:2606.15524v1 Announce Type: new Abstract: In this work, we present a simple, exact, and fully analytical method for generating light sheets parallel to the propagation direction, with amplitude and phase envelopes structured on demand. We validate the approach theoretically and experimentally by imprinting images onto light sheets, and we compare the theoretical performance with that obtained using an alternative strategy based on arrays of Frozen Waves (FWs). In this context, the proposed method provides a more direct and flexible control of the field envelopes on the light sheets, resulting in higher-fidelity reconstructions than those achieved with FW-based approaches. The method thus offers a versatile framework for structured light-sheet generation, with potential applications in optical manipulation, microscopy, and 3D holographic imaging.
arXiv:2606.16054v1 Announce Type: new Abstract: Research on artificial intelligence and democracy has grown quickly over the last decade. A shared conclusion in this literature is that AI does not create new democratic problems so much as it makes old ones worse. We now see this across information ecosystems, in elections, and in public administration. However, despite growing evidence, we lack a clear way to prioritize risks in this area, compare them across domains, and identify where democratic control is most likely to break down. So, our problem is: How can we systematize the problems that AI systems pose to democratic processes? This paper argues that principal agent theory may fit the task. In many phases of democratic systems, principals delegate key functions to AI systems and their providers without really being able to monitor how these systems operate or the outputs they produce. Treating AI as a delegation problem helps identify accountability gaps and other governance failures. Most importantly, as we shall illustrate, it provides metrics for empirical assessments of AI impact on democracy. As a second analytical element, we draw on the NIST AI Risk Management Framework and its seven characteristics of trustworthy AI, which supply substantive criteria for evaluating delegated tasks. Operationalized across the three domains through measurable indicators and domain specific trustworthiness criteria, we propose an analytical framework that centers on institutional assessability as the central condition for democratic control over AI. However, we stress that how severe a harm is, and how much risk is acceptable, are evaluative judgments that current methodologies neither acknowledge nor operationalize. This becomes acute when such evaluative judgments are (silently) delegated to private vendors. We identify this as a strong limitation left for future work.
arXiv:2606.16126v1 Announce Type: new Abstract: The frictional force arising from the interaction of ions with a magnetized electron beam in a strong magnetic field is investigated. The problem is reduced to the interaction of quasiparticles, or Larmor circles, with ions. A Hamiltonian describing this interaction is derived. The role of various impact parameters in the process of ion cooling is elucidated. The cases of both positively and negatively charged ions are considered.
arXiv:2606.15527v1 Announce Type: new Abstract: Typical video object-centric learning (VOCL) approaches employ slot-based frameworks that rely on reconstruction-driven encoder-decoder architectures, where learning is mediated by two spatial maps: attention maps from the encoder and object maps from the decoder. As these two distinct maps exhibit different properties, a recent dense alignment strategy attempted to reconcile this discrepancy by enforcing agreement across all spatio-temporal patches via contrastive learning. However, this indiscriminate alignment inadvertently propagates the inherent weaknesses of each module, such as noisy encoder predictions and blurred decoder boundaries. Moreover, computing dense similarities across all pairs incurs a computational cost quadratic in the total number of spatio-temporal patches, severely limiting scalability. Motivated by this, we propose Selective Synergistic Learning (SSync). Instead of exhaustive patch-to-patch alignment, SSync prevents error propagation by selectively distilling only the most reliable cues: leveraging the encoder strictly for boundary refinement and the decoder for interior denoising. This is realized via a pseudo-labeling with linear complexity, eliminating the need for quadratic spatial comparisons. Also, to prevent the reinforcement of architectural biases like slot redundancy, we introduce a transitive pseudo-label merging that consolidates overlapping slots based on spatio-temporal activation consistency. Extensive studies demonstrate that SSync improves decomposition quality and serves as a versatile, plug-and-play module while also exhibiting exceptional robustness to slot configurations. Code is available at github.com/wjun0830/SSync.
arXiv:2606.15528v1 Announce Type: new Abstract: Assistive mobility devices support independence for people with lower-limb disabilities, yet many are designed and evaluated in clinical or controlled environments. In Indian households, narrow spaces and dense furniture often make assistive devices difficult to use indoors, leading people to rely on improvised movements or support from family members and caregivers. In this work, we explore domestic mobility through a participatory and co-speculative design approach, focusing on how people with lower-limb disabilities navigate and maneuver within their homes. We conducted a series of semi-structured interviews and bespoke booklet-based participatory workshops with 22 participants with lower-limb impairments. To support reflection and discussion, we designed bilingual bespoke booklets grounded in domestic design frictions, using images and scenarios to encourage storytelling, critique, and speculation. Our findings reveal mobility challenges that differ significantly from those typically observed in clinical contexts. Rather than yielding a fixed set of design solutions, the study contributes situated insights into domestic mobility frictions, participant articulation, and the limits of speculative participation in this context.
arXiv:2606.15733v1 Announce Type: new Abstract: Instruction-tuned language models can answer the same causal-reasoning question differently after its English variable names are replaced by type-preserving placeholders, although the structural causal model and the gold answer are unchanged. We ask whether this lexical gap reflects information loss in the placeholder view or a misaligned read-out from a representation that still carries answer-relevant content. Vernier uses a paired-view weight update as an instrument and then inspects the mechanism left after the gap closes. In the working regimes, the evidence favours representational misalignment. A variable-name probe becomes more accurate on the placeholder view, and activation patching on Qwen-7B, Qwen-14B, and Llama-3.1-8B shows that the decision-token representation can transfer answer identity between views. The update that realigns the views is counterfactual augmentation over original and placeholder prompts, while the answer-subspace KL mainly sharpens intermediate answer-belief agreement. Success is bounded by model family, scale, and task. CRASS transfer is reliable across Qwen scales and Llama, e-CARE remains weak, and preliminary non-causal rename tasks show a similar qualitative pattern.
arXiv:2606.15532v1 Announce Type: new Abstract: Emotional intelligence (EI) in Large Language Models (LLMs) is often evaluated through static understanding tasks or single-response dialogue generation. However, emotion management is interactive: a good model should not only recognize a user's emotion, but also improve the user's emotional and relational state over several turns. We introduce EIBench, a simulator-based benchmark for interactive emotion management. EIBench contains 2,222 scenarios, with 2,009 for training and 213 for held-out testing. The scenarios are organized by a 2x2 taxonomy covering Support, Defense, Repair, and Charm, which together capture different forms of support, boundary maintenance, trust repair, and rapport building. In each scenario, an LLM simulator plays the user, updates an emotion-relation state after each turn, and maps the final state to an anchor-based score. This design makes EIBench both an evaluation benchmark and a training environment: the final state gives the outcome reward, while the per-turn state updates provide dense feedback for RL. We evaluate 15 open- and closed-source LLMs. Current models perform well on support and rapport-building scenes, but struggle with boundary maintenance under user pressure. To improve the EI ability of LLMs, we propose Centered Turn-Credit GRPO (CTC-GRPO), a GRPO extension that reuses the simulator's per-turn state updates as dense turn-level feedback while preserving the final outcome reward. CTC-GRPO improves Qwen3-8B from -22.4 to +22.4 on EIBench and also improves on out-of-distribution evaluations including SAGE (+12.4) and EQBench3 (+20.9%). Our results show that simulator-tracked user states can support both evaluation and training for multi-turn emotion management.
arXiv:2606.15534v1 Announce Type: new Abstract: Re-rendering an existing video from a novel camera viewpoint requires the output to follow the prescribed camera trajectory while preserving the appearance and dynamics of the original scene across every frame. Existing methods rely on per-frame pose embeddings, noisy point-cloud renderings, or implicit learned correspondences, none of which provides an explicit, temporally continuous link between source and target pixels. We propose Track2View, which conditions a video diffusion transformer on paired 3D point tracks: sparse trajectories of scene points projected into both the source and target camera views. These tracks provide explicit spatiotemporal correspondences that are temporally continuous by construction, encoding what content should appear where and when. At the core of Track2View is a dual-view track conditioner that transfers visual context from source to target view through parameter-free geometric operations and learned temporal aggregation, ensuring generalization to arbitrary camera trajectories without memorizing specific motions. We further introduce a data curation pipeline that extracts one-to-one track correspondences by running a 3D point tracker on temporally concatenated multi-camera view pairs. On a 400-video benchmark spanning static and dynamic scenes, Track2View achieves state-of-the-art results across visual quality, view synchronization, and camera accuracy, reducing rotation error by 30-65% and translation error by 61-72% relative to leading baselines. Project page is available at this https URL: https://qjizhi.github.io/track2view
arXiv:2606.15536v1 Announce Type: new Abstract: Ongoing discussions in Human-Computer Interaction(HCI) have examined the role of AI-based tools in health information seeking, particularly within sensitive domains such as reproductive health. We introduce "OpenBloom," a web application and an exploratory design probe that utilizes Large Language Models (LLMs) to turn reproductive health articles into question-based prompts to explore stigma around reproductive wellbeing. Through a survey study with 34 participants across their 136 interactions with OpenBloom, we explore how AI-generated question-based learning interacts with sociocultural stigma, contextual sensitivity, and reflexiveness. While current LLM outputs largely meet expectations for non-offensiveness, they default to superficial rephrasing or factual recall and lack critical reflections. We discuss implications for applying Feminist HCI, contestability, and value-sensitive AI frameworks to future LLM-mediated reproductive health technologies.
arXiv:2606.15540v1 Announce Type: new Abstract: Pathological speech from patients with neurodegenerative and neuromotor disorders is often acoustically distorted and linguistically fragmented, making pathological speech reconstruction necessary to recover intended textual content from distorted and incomplete speech recordings. Crucially, such recordings are rarely uniformly degraded: some words or short phrases remain reliable and can serve as audible anchors for reconstructing the corrupted surrounding content. We introduce Anchor-gated Phonetic Group Relative Policy Optimization (AP-GRPO), a GRPO framework with phonetic reward that aligns speech language models (SLMs) through audible-anchor preservation and inter-anchor phonetic compatibility to the original speech signal. AP-GRPO consists of: (i) an anchor-gated reward that matches reliable audible anchors in clear regions; and (ii) an inter-anchor phonetic alignment reward that evaluates whether recovered contents are phonetically supported by the corresponding corrupted inter-anchor speech span. Across four disease conditions, AP-GRPO improves faithful speech reconstruction, and the learned anchor constraint automatically adapts to each condition and thus reveals interpretable disease-specific profiles: conditions with severe articulatory degradation require stronger anchor enforcement, whereas milder impairment or linguistically impaired conditions rely more on phonetic alignment for inter-anchor recovery.
arXiv:2606.15549v1 Announce Type: new Abstract: The adoption of AI agents is increasing rapidly. Terminal AI agents, i.e., AI agents that run in terminal environments, are a widely used type of AI agents. Terminal AI agents rely heavily on shell command execution to interact with the host systems. They adopt a three-list command-gating mechanism to mitigate security risks introduced by command execution, with denylists serving as the load-bearing component. However, modern operating systems often ship a large, ever-expanding set of shell commands with complex functionalities. Our observation is that even a built-in denylist of Claude Code, well-maintained by its developers, can overlook bypass commands that invalidate its effectiveness. Such negligence leads to fragile command denylists that cannot even block operations that practitioners expect them to block. This paper presents the first systematic characterization of command denylist fragility in terminal AI agents. The paper formalizes the command denylist fragility problem and proposes an LLM-driven pipeline, CmdNeedle, to detect such fragility. It prompts the LLM to propose possible bypasses and iteratively repairs them using feedback from a validator that executes them in a sandbox. In the evaluation, we applied CmdNeedle to 1,709 real-world command denylists (containing 13,332 denylist rules) collected from GitHub. The evaluation shows several key findings, including that 69.0--98.6% of the denylists are fragile, that this fragility occurs consistently across projects and agents, and the validity of several possible root causes for this fragility. Our pipeline and findings will hopefully facilitate future research and practice regarding the command denylists used by AI agents.