Academic Intelligence · Curated Daily

探索全球前沿学术脉络

AcademicHub 汇聚顶级期刊与预印本平台的实时文献。定制您的专属科研雷达,利用大语言模型自动生成交叉领域文献分析简报。

01.
arXiv (CS.LG) 2026-06-17

Manifold GCN: Diffusion-based Convolutional Neural Network for Manifold-valued Graphs

arXiv:2401.14381v3 Announce Type: replace Abstract: We propose two graph neural network layers for graphs with features in a Riemannian manifold. First, based on a manifold-valued graph diffusion equation, we construct a diffusion layer that can be applied to an arbitrary number of nodes and graph connectivity patterns. Second, we model a tangent multilayer perceptron by transferring ideas from the vector neuron framework to our general setting. Both layers are equivariant under node permutations and the feature manifold's isometries. These properties have led to a beneficial inductive bias in many deep-learning tasks. Furthermore, they enable novel, more flexible feature designs. Numerical examples on synthetic data and an Alzheimer's classification application on triangle meshes of the right hippocampus demonstrate the usefulness of our new layers: While they apply to a much broader class of problems, they outperform task-specific state-of-the-art networks.

03.
arXiv (CS.CL) 2026-06-12

How Fine-Grained Should a RAG Benchmark Be? A Hierarchical Framework for Synthetic Question Generation

Evaluating retrieval-augmented generation (RAG) systems requires benchmarks that capture diverse question characteristics, yet practitioners lack empirical guidance on which dimensions to vary and at what granularity. We present HieraRAG, a hierarchical framework for studying granularity in RAG benchmark construction, defining optimal granularity as the level that maximizes discriminative power (the standard deviation of generation quality across categories) within a given RAG configuration. As a case study, we generate 5,872 synthetic question-answer (QA) pairs from FineWeb-10BT across 3 dimensions (Question Complexity, Answer Type, Linguistic Variation) at 3 granularity levels (2, 4, and 8 categories). With a BM25+Falcon-3-10B pipeline, optimal granularity varies by dimension: complexity benefits from fine-grained distinctions (discriminative power: 0.053) while answer type and linguistic variation peak at medium granularity. We introduce a Coherence Ratio metric to quantify whether fine-grained splits cleanly subdivide parent categories, revealing structural differences across dimensions (Question Complexity: 0.40 vs. Answer Type: 1.44). Human evaluation of 110 stratified QA pairs confirms synthetic quality. While these specific findings reflect a single configuration, HieraRAG provides a portable procedure and validation metric for practitioners to determine evaluation granularity within their own RAG settings.

04.
arXiv (CS.LG) 2026-06-19

Global Convergence of Gradient Descent for Score Matching in Gaussian Mixtures via Reverse Fisher Divergence

arXiv:2606.19876v1 Announce Type: new Abstract: The score matching problem is a central training objective in modern generative modeling, diffusion models, fitting unnormalized statistical models, and inverse problems. A standard approach is to minimize the forward Fisher divergence, where the expectation is taken with respect to the teacher distribution. However, recent results show that even in simple Gaussian mixture model settings, this objective can lead to undesirable and initialization-dependent convergence behavior. In this paper, we study an alternative objective: the reverse Fisher divergence, where the expectation is taken with respect to the student distribution. We analyze gradient descent (GD) for fitting Gaussian mixture models and show that this change in the objective leads to significantly better optimization properties. First, when the teacher distribution is a single Gaussian and the student is a Gaussian mixture model with fixed weights and identity covariances, we prove the global convergence of GD from arbitrary initializations. Second, we extend the analysis to the case where the teacher is also a Gaussian mixture model and prove global convergence guarantees under a global random initialization scheme and a $\widetilde{\Omega}(1)$-separation assumption on the target means. In particular, with high probability, each student component converges near its closest teacher component, and we provide conditions under which the student distribution converges in total variation distance. Our proofs rely on a new Lyapunov-based analysis of the gradient descent dynamics, showing that the reverse Fisher divergence has a much more favorable optimization landscape than the forward Fisher divergence.

05.
arXiv (CS.AI) 2026-06-11

Information bottleneck for learning the phase space of dynamics from high-dimensional experimental data

arXiv:2604.24662v2 Announce Type: replace-cross Abstract: Identifying the dynamical state variables of a system from high-dimensional observations is a central problem across physical sciences. The challenge is that the state variables are not directly observable and must be inferred from raw high-dimensional data without supervision. Here we introduce DySIB (Dynamical Symmetric Information Bottleneck) as a method to learn low-dimensional representations of time-series data by maximizing predictive mutual information between past and future observation windows while penalizing representation complexity. This objective operates entirely in latent space and avoids reconstruction of the observations. We apply DySIB to an experimental video dataset of a physical pendulum, where the underlying state space is known. The method, with hyperparameters of the learning architecture set self-consistently by the data, recovers a two-dimensional representation that matches the dimensionality, topology, and geometry of the pendulum phase space, with the learned coordinates aligning smoothly with the canonical angle and angular velocity. These results demonstrate, on a well-characterized experimental system, that predictive information in latent space can be used to recover interpretable dynamical coordinates directly from high-dimensional data.

06.
arXiv (CS.AI) 2026-06-19

What Do Safety-Aligned LLMs Learn From Mixed Compliance Demonstrations?

arXiv:2606.20508v1 Announce Type: new Abstract: Prior work has shown that in-context demonstrations can jailbreak language models, but it remains unclear how models interpret different types of compliance demonstrations. We study this by mixing benign compliance demonstrations (non-harmful request, helpful response) with harmful compliance demonstrations (harmful request, helpful response) and testing three hypotheses about how demonstration composition drives harmful compliance. Across four models, we find that benign and harmful demonstrations are not interchangeable: benign demonstrations can either reduce or increase harmful compliance depending on the model. We further show that preference optimization is the critical training stage that prevents benign demonstrations from increasing harmful compliance, that demonstration ordering exhibits strong recency bias, and that models differ in how refusal interacts with in-context learning: some adopt demonstrated formatting even when refusing, while others override all in-context signals upon refusal. Taken together, this work moves beyond showing that demonstration-based jailbreaking works to characterizing how it works: what models extract from compliance demonstrations depends on demonstration content, ordering, and training methodology.

07.
arXiv (CS.AI) 2026-06-12

Valid Inference with Synthetic Data via Task Exchangeability

arXiv:2606.13629v1 Announce Type: cross Abstract: There is a proliferation of work arguing for the use of synthetic data in scientific research. For example, social scientists are arguing for the use of LLM-generated "silicon samples" in pilot studies; AI evaluations increasingly rely on "LLM-as-a-judge" outputs; and proteomics research is accelerated by generative models that produce synthetic protein structures. These developments raise an intriguing possibility: synthetic data may help researchers ask more questions, run more studies, and accelerate discovery. But they also raise a fundamental concern: synthetic data can be biased, noisy, and misspecified. In this work, we propose statistical principles for using synthetic data in scientific research with provable validity guarantees. The key insight is a new technical condition that we call task exchangeability. Informally, this is a requirement that the researcher can identify historical tasks, for which real data is available, such that their current task of interest is exchangeable with the historical tasks in an appropriate mathematical sense. We develop methods for valid inference under task exchangeability, together with extensions that provide guarantees even beyond exchangeability. We demonstrate the framework on public opinion surveys with silicon samples and AI evaluation with autoraters.

08.
arXiv (CS.LG) 2026-06-12

TEDD: Robust Detection of Unstable Temporal Features

arXiv:2606.12643v1 Announce Type: new Abstract: When working with real-world temporal data, it is common to encounter features whose distribution is changing over time. The naive employment of Machine Learning models on this unstable data might lead to rapidly degrading performance, especially if the new distribution is much different from what was previously seen during training. In order to cope with this problem, it is critical to automatically identify features that are changing over time. With these features detected, data scientists and other practitioners will be able to mitigate the issue (for instance, by applying data transformations), deploying more robust models that retain high performance for longer periods of time. In this paper, we describe which temporal changes a feature should not suffer from, and propose TEDD, a technique to a) identify when a dataset might lead to an unstable Machine Learning model and b) automatically detect which features cause such lack of robustness. In order to achieve it, we leverage a regression model to highlight which features contribute to a good prediction of an instance's timestamp. We compare our approach to other methods in real and synthetic data, testing their detection capability on all simple change patterns. We show that our method: detects all types of basic changes, both for numerical and categorical features; can detect multivariate drifts; returns a comparable value measuring the amount of change of each feature; requires no parameter tuning; and is scalable both on number of features and instances of the dataset.

09.
arXiv (quant-ph) 2026-06-11

Holographic Complexity, Extremality, and Cosmic Censorship

arXiv:2604.20170v2 Announce Type: replace-cross Abstract: We propose a holographic complexity origin for the third law of black-hole mechanics and weak cosmic censorship. In both complexity equals action and complexity equals volume prescriptions, the relative complexity between subextremal and extremal AdS black holes diverges logarithmically. For overcharged RN-AdS, explicit calculations in both prescriptions show that the near-singularity action terms are power-law divergent or finite, while the maximal-volume contribution is finite. Thus, the extremal-to-naked relative complexity also diverges, obstructing finite-time transitions.

10.
Nature (Science) 2026-06-17

A prototype differential atom interferometer for fundamental physics

Gravitational waves and ultralight dark matter are among the most compelling frontiers in fundamental physics, motivating proposals for very-long-baseline atom interferometerssuch as AION1, MAGIS2, AICE3 and AEDGE4 that aim to detect at frequencies at which ground-based5 and space-borne6 laser interferometers lose sensitivity. Very-long-baseline atom interferometers look for signals by comparing the quantum phase evolution of widely separated atomic ensembles interrogated by a common laser. However, their performance depends critically on suppressing noise sources, particularly laser phase noise. The experimental validation of such noise rejection remains an important challenge. Here we demonstrate a prototype differential atom interferometer based on the single-photon clock transition of fermionic 87Sr. Thus, we obtain a gradiometer configuration with a species intrinsically suited to kilometre-scale and space-baseline operation. The instrument operates at the standard quantum limit7 with no excess noise beyond atom shot noise. The differential configuration maintains quantum-limited sensitivity in the presence of several radians of artificially injected laser phase noise per shot, which emulates the conditions expected in a very-long-baseline atom interferometer. We also demonstrate the recovery of coherent oscillatory signals across a broad frequency range under fully phase-randomized conditions, a capability that is inaccessible to a single interferometer operating in the same regime. These results provide an experimental validation of the noise-immune measurement principle underlying very-long-baseline atom interferometers and mark an important step towards next-generation quantum sensors for gravitational-wave detection and searches for ultralight dark matter8,9. A prototype differential atom interferometer operates at the standard quantum limit with no excess noise beyond atom shot noise, achieving performance in line with the specifications for future long-baseline atom interferometers.

11.
arXiv (CS.AI) 2026-06-18

Generating Natural and Expressive Robot Gestures through Iterative Reinforcement Learning with Human Feedback using LLMs

arXiv:2606.18747v1 Announce Type: cross Abstract: Expressive gestures are essential for natural and effective communication, complementing speech when verbal cues alone are insufficient (e.g., pointing). For social robots such as the humanoid Pepper, producing natural and expressive movements is critical for improving human-robot interaction (HRI) and long-term acceptance. However, generating gestures remains challenging due to reliance on expert-authored animations, resulting in rigid behaviors that are impractical for dynamic and diverse environments. Alternatively, machine learning approaches often struggle to capture perceived naturalness, becoming increasingly challenging with more degrees of freedom. Consequently, producing expressive robot gestures requires a system that can adapt to the environment while adhering to social norms and physical constraints. Recent advances in large language models (LLMs) enable dynamic code generation, offering new opportunities for runtime gesture synthesis from natural language. In this paper, we integrate ChatGPT into the humanoid robot Pepper to generate co-speech gestures aligned with conversational output. While this baseline enables flexible gesture generation, the resulting motions are often perceived as stiff and unnatural. To address this limitation, we introduce an iterative reinforcement learning with human feedback (RLHF) system that finetunes gesture generation based on user evaluations, leveraging an iterative user study to compare Pepper's generated gestures. Our results show that RLHF improved the LLM's co-speech generative capabilities, producing more expressive, relevant and fluid movements.

12.
arXiv (CS.CL) 2026-06-18

Learning User Simulators with Turing Rewards

Learning to simulate human users in interactive settings could advance the training of agent assistants, evaluation of personalization systems, research in the social sciences, and more. Existing approaches generally do so by training a large language model (LLM) to match a single ground truth response, either by maximizing the log probability or by using a similarity reward. We instead propose {Turing-RL}: a Turing-Test-based reinforcement learning approach for training user simulator models. {Turing-RL} uses a discriminative Turing reward with an LLM judge to score how indistinguishable a generated response is from the real user's given the user's history, and the user simulator LLM learns to produce responses indistinguishable from what the user could have said with such rewards. Across two different domains–conversational chat and Reddit forum discussion–we find that {Turing-RL} consistently outperforms baseline methods on both LLM and human evaluation metrics. Our study suggests that optimizing for indistinguishability, rather than response matching, is effective for learning user simulators.

13.
arXiv (CS.CL) 2026-06-12

How reliable are LLMs when it comes to playing dice?

We investigate the probabilistic reasoning capabilities of large language models through a controlled benchmarking study on discrete probability problems. We constructed two datasets, respectively a set of standard exercises and a set of counterintuitive exercises, designed to trigger heuristic reasoning, and evaluated 8 state-of-the-art models, each tested with and without Chain-of-Thought prompting. Models achieve an average accuracy of 0.96 on standard problems but only 0.59 on counterintuitive ones. We further provide empirical evidence of token bias: performance drops by over 20% when canonical formulations are replaced by disguised variants. Embedding misleading suggestions in the prompt reduces performance by up to 34%, with no model proving immune. Taken together, the reported findings suggest that current LLMs are not yet genuine probabilistic reasoners, despite their success in advanced mathematical problems.

14.
arXiv (CS.CV) 2026-06-16

MIRAGE: Runtime Scheduling for Multi-Vector Image Retrieval with Hierarchical Decomposition

To effectively leverage user-specific data, retrieval augmented generation (RAG) is employed in multimodal large language model (MLLM) applications. However, conventional retrieval approaches often suffer from limited retrieval accuracy. Recent advances in multi-vector retrieval (MVR) improve accuracy by decomposing queries and matching against segmented images. They still suffer from sub-optimal accuracy and efficiency, overlooking alignment between the query and varying image objects and redundant fine-grained image segments. In this work, we present an efficient scheduling framework for image retrieval - MIRAGE. First, we introduce a novel hierarchical paradigm, employing multiple intermediate granularities for varying image objects to enhance alignment. Second, we minimize redundancy in retrieval by leveraging cross-hierarchy similarity consistency and hierarchy sparsity to minimize unnecessary matching computation. Furthermore, we configure parameters for each dataset automatically for practicality across diverse scenarios. Our empirical study shows that, MIRAGE not only achieves substantial accuracy improvements but also reduces computation by up to 3.5 times over the existing MVR system.

15.
arXiv (CS.CL) 2026-06-15

Fusing Stylometric and Embedding Systems to Estimate Authorship Likelihood Ratios in Japanese

The likelihood ratio framework is widely recognized as the logically and legally sound basis for evidential analysis across forensic sciences, and its importance is increasingly acknowledged in analyses of authorship in textual evidence. To date, however, its application has been confined to English-language texts. Meanwhile, authorship attribution has traditionally relied on a diverse array of stylometric features, even as the rise of pre-trained large language models enables new contextual-embedding approaches. Combining these diverse approaches through fusion promises enhanced performance, yet it has not been applied to integrate stylometric-feature systems with embedding-based systems within the likelihood ratio paradigm. This study is the first to apply likelihood ratio-based forensic text comparison to Japanese digital texts, using ~1,000-character excerpts from blogs, to 1) evaluate system performance and likelihood ratio magnitudes and 2) assess the impact of fusing stylometric-feature systems with embedding-based systems. The results demonstrate that the fused system maintains excellent calibration while 1) increasing consistent-with-fact likelihood ratio magnitudes; 2) decreasing contrary-to-fact likelihood ratio magnitudes and 3) improving overall discriminability. The best-performing fusion achieved a log-likelihood-ratio cost of 0.32484, illustrating both the feasibility of likelihood ratio framework for Japanese and the benefits of fusion across heterogeneous systems.

16.
arXiv (CS.LG) 2026-06-19

HGCN(O): A Self-Tuning GCN HyperModel Toolkit for Outcome Prediction in Event-Sequence Data

arXiv:2507.22524v3 Announce Type: replace Abstract: We propose HGCN(O), a self-tuning toolkit using Graph Convolutional Network (GCN) models for event sequence prediction. Featuring four GCN architectures (O-GCN, T-GCN, TP-GCN, TE-GCN) across the GCNConv and GraphConv layers, our toolkit integrates multiple graph representations of event sequences with different choices of node- and graph-level attributes and in temporal dependencies via edge weights, optimising prediction accuracy and stability for balanced and unbalanced datasets. Extensive experiments show that GCNConv models excel on unbalanced data, while all models perform consistently on balanced data. Experiments also confirm the superior performance of HGCN(O) over traditional approaches. Applications include Predictive Business Process Monitoring (PBPM), which predicts future events or states of a business process based on event logs.

17.
arXiv (CS.AI) 2026-06-19

Playful Agentic Robot Learning

arXiv:2606.19419v1 Announce Type: cross Abstract: Current agentic robot systems can write executable Code-as-Policy programs, observe feedback, and revise behavior across multiple attempts, but they remain largely task-driven: reusable skills are acquired only after explicit instructions. We study Playful Agentic Robot Learning, where an embodied coding agent uses self-directed play as a continual skill-learning stage before downstream tasks arrive. We introduce RATs, Robotics Agent Teams designed for play-time skill acquisition. During play, RATs proposes novel yet learnable exploratory tasks, plans and executes robot-code policies, verifies intermediate progress, diagnoses failures, retries with dense, step-level feedback, and distills successful executions into a persistent code skill library. At test time, the agent reuses relevant skills from this frozen library to help solve new tasks. Experiments in LIBERO-PRO and MolmoSpaces show that play-learned skills improve held-out downstream tasks over no-play and random-play baselines, with 20.6 and 17.0 percentage-point gains over CaP-Agent0 on LIBERO-PRO and MolmoSpaces, respectively. Moreover, the learned skills can be plugged into other inference-time Code-as-Policy agents by simply retrieving them into the context, improving RoboSuite and real-world transfer by 8.9 and 8.8 points, respectively, without finetuning the underlying model.

18.
arXiv (CS.LG) 2026-06-18

RouteJudge: An Open Platform for Reproducible and Preference-Aware LLM Routing

arXiv:2606.18774v1 Announce Type: new Abstract: We present RouteJudge, an online pairwise preference evaluation framework for LLM routing systems, with a public platform available at https://routejudge.cn. Different from model-level response evaluation, RouteJudge focuses on router-level decision quality. For each user query, multiple routing strategies independently recommend candidate models under the same model pool and budget constraints. The selected model responses are then presented to users through anonymous pairwise comparisons, and the resulting user preferences are attributed back to the routing strategies behind the compared responses. Each evaluation record stores the query, routing decisions, model responses, preference labels, cost, latency, and task metadata, enabling preference-aware, cost-aware, and task-conditioned analysis of LLM routers. To support the continuous expansion of routing methods in RouteJudge, we further release ORBIT (Optimal Routing and Budgeted Inference Toolbox), a modular and extensible toolbox that standardizes the end-to-end workflow of LLM routing. ORBIT provides unified interfaces for benchmark loading, query representation, router implementation, budget-aware evaluation, and method comparison, allowing researchers to develop and evaluate routing algorithms under consistent protocols. It also serves as the submission and integration layer for RouteJudge: researchers can implement routing methods within ORBIT, validate them on existing routing benchmarks, and submit compatible routers for online preference-based evaluation. The code of ORBIT is available at https://github.com/AIGNLAI/LAMDA-ORBIT.

19.
arXiv (CS.LG) 2026-06-19

CAGE: Curvature-Aware Gradient Estimation For Accurate Quantization-Aware Training

arXiv:2510.18784v3 Announce Type: replace Abstract: Despite significant work on low-bit quantization-aware training (QAT), there is still an accuracy gap between such techniques and native training. To address this, we introduce CAGE (Curvature-Aware Gradient Estimation), a new QAT method that augments the straight-through estimator (STE) gradient with a curvature-aware correction designed to counteract the loss increase induced by quantization. CAGE is derived from a multi-objective view of QAT that balances loss minimization with the quantization constraints, yielding a principled correction term that depends on local curvature information. On the theoretical side, we introduce the notion of Pareto-optimal solutions for quantized optimization, and establish that CAGE yields strong convergence guarantees in the smooth non-convex setting. In terms of implementation, our approach is optimizer-agnostic, but we provide a highly-efficient implementation that leverages Adam statistics. CAGE significantly improves upon the prior state-of-the-art methods in terms of accuracy, for similar computational cost: for QAT fine-tuning, it halves the compression accuracy loss relative to the prior best method, while for QAT pre-training of Llama models, its accuracy for 3-bit weights-and-activations (W3A3) matches the accuracy achieved at 4-bits (W4A4) with the prior best method. The official implementation can be found over https://github.com/IST-DASLab/CAGE .

20.
arXiv (CS.CV) 2026-06-15

Efficient Online 3D Multi-Camera Multi-Object Tracking and Pose Estimation

This paper proposes a fast and online method for jointly performing 3D multi-object tracking and pose estimation using multiple monocular cameras. Our algorithm requires only 2D bounding box and pose detections, eliminating the need for costly 3D training data or computationally expensive deep learning models. Our solution is an efficient implementation of a Bayes-optimal multi-object tracking filter, enhancing computational efficiency while maintaining accuracy. We demonstrate that our algorithm is significantly faster than state-of-the-art methods without compromising accuracy, using only publicly available pre-trained 2D detection models. We also illustrate the robust performance of our algorithm in scenarios where multiple cameras are intermittently disconnected or reconnected during operation.

21.
arXiv (math.PR) 2026-06-15

Laws of Large Numbers for Non-Independent Random Variables on Hyperspaces with respect to the Hausdorff Metric

arXiv:2011.07199v5 Announce Type: replace Abstract: This paper investigates the limit behavior of the Minkowski sums for sequences of set-valued random variables. When the underlying space is finite dimensional, by using the support function, we establish the weak and strong laws of large numbers for non-independent random variables in the hyperspace with respect to the Hausdorff metric $d_H$.

22.
arXiv (CS.CV) 2026-06-16

Robust Spoofed Speech Detection via Temporal Pyramid Modeling

Spoofed speech detection is increasingly challenged by realistic synthesis, voice conversion, and replay attacks, with cross-dataset generalization remaining a major limitation. This work we propose a Temporal Pyramid Adapter that utilize parallel temporal convolutions with varying receptive fields to capture multi-scale spoofing cues, ranging from local artifacts to global prosodic irregularities. We also integrated self-supervised XLS-R representations combined with front-end adapters, including Mel, Sinc, and a Temporal Pyramid design for multi-scale temporal modeling. The proposed model is evaluated cross multiple benchmark including ASVspoof 2017, ASVspoof 2021 (DF/LA), PartialSpoof, DiffSSD, and multilingual HQ-MPSD datasets. Experimental results demonstrate that Temporal Pyramid model obtained AUC of 99.24% and a EER of 3.87% on the PartialSpoof database, which is significantly outperforming the base model and several SOTA baseline such as LCNN-BLSTM (9.87% EER) and TRACE (8.08% EER). Additionally, multilingual evaluations confirm that while spoofing artifact are independent from language. While self-supervised representations improve robustness, performance degrades under domain and language shifts, highlighting the need for better adaptation and calibration strategies.

23.
arXiv (CS.CL) 2026-06-19

MENTOR: Reinforcement Learning via Flexible Teacher-Optimized Rewards for Tool-Use Distillation

Distilling the tool-use capabilities of large language models (LLMs) into small language models (SLMs) is essential for their practical application. The predominant approach, supervised fine-tuning (SFT), suffers from poor out-of-domain (OOD) generalization due to its rigid alignment with static teacher trajectories. While reinforcement learning (RL) offers an alternative, the capacity limitations of SLMs pose a severe dilemma: sparse outcome rewards provide insufficient guidance, whereas strict trajectory matching imposes overly restrictive constraints. To bridge this capacity-driven gap, we propose MENTOR, which introduces a flexible yet process-aware reward structure. Instead of enforcing rigid replication, MENTOR uses the teacher's reference to guide tool-use behavior, balancing behavioral alignment with downstream performance. Extensive experiments on controlled executable-tool benchmarks demonstrate that MENTOR improves OOD tool-use performance compared to SFT and strict RL baselines. Our findings suggest that within verifiable tool-use environments, flexible tool-use alignment offers a more effective approach than strict trajectory replication for developing adaptable small models.

24.
arXiv (CS.AI) 2026-06-19

The Autonomy Tax: Defense Training Breaks LLM Agents

arXiv:2603.19423v2 Announce Type: replace-cross Abstract: Large language model (LLM) agents increasingly rely on external tools (file operations, API calls, database transactions) to autonomously complete complex multi-step tasks. Practitioners deploy defense-trained models to protect against prompt injection attacks that manipulate agent behavior through malicious observations or retrieved content. We reveal a fundamental capability-alignment paradox: defense training designed to improve safety systematically destroys agent competence while failing to prevent sophisticated attacks. Evaluating defended models against undefended baselines across 97 agent tasks and 1,000 adversarial prompts, we uncover three systematic biases unique to multi-step agents. Agent incompetence bias manifests as immediate tool execution breakdown, with models refusing or generating invalid actions on benign tasks before observing any external content. Cascade amplification bias causes early failures to propagate through retry loops, pushing defended models to timeout on 99\% of tasks compared to 13\% for baselines. Trigger bias leads to paradoxical security degradation where defended models perform worse than undefended baselines while straightforward attacks bypass defenses at high rates. Root cause analysis reveals these biases stem from shortcut learning: models overfit to surface attack patterns rather than semantic threat understanding, evidenced by extreme variance in defense effectiveness across attack categories. Our findings demonstrate that current defense paradigms optimize for single-turn refusal benchmarks while rendering multi-step agents fundamentally unreliable, necessitating new approaches that preserve tool execution competence under adversarial conditions.

25.
arXiv (CS.CL) 2026-06-16

Understanding Scam Trends and Rail Paths from Reddit Self-Disclosure Narratives

Online scam behavior is inherently multi-stage, and the lifecycle includes temporally ordered rails and events rather than isolated signals. Existing works analyze characteristics of scam types and rails, but they do not track scam trends across years. Moreover, the work on the relations between rails is hampered due to the lack of open-source datasets with annotations and coverage of different scam types. To address these gaps, we build a dataset to analyze the yearly trend of scam characteristics and rail paths using Reddit self-disclosure narratives from 2023 to 2025. We collect 21,304 posts from scam-related subreddits with at least one rail among identity, communication, platform, and payment for trend analysis by heuristic annotation. Then, we label 1,800 posts containing explicit or recoverable scam chains by an LLM-assisted method for scam path analysis. The method is evaluated with human annotation. Lastly, we run a topic model on the comments of the posts to analyze the community support behavior. The results reveal that scam processes are predominantly multi-rail. Across years, different scam types and rail components dominate. Different scam types vary systematically in path complexity. Reddit support behaviors have become more detailed over time. This work supports synthetic scam chain data simulation and AI-related scam risk assessment, though findings may not generalise to other platforms.