AI conference papers

AI conference paper library

Search papers from 30+ top conferences, including NeurIPS, ICLR, ICML, CVPR, ACL and AAAI. Filter by keyword, then open any paper for its abstract, authors and links.

arXiv

Extremal Chowla sets and their linear analogues: A human-AI mathematical investigation using Co-Scientist

Mohsen Aliabadi, Keith Driscoll, Elliot Krop, Petar Sirkovic, Everett Sullivan, Elahe Vedadi

We introduce an extremal invariant associated with Chowla-type order conditions in finite groups. A nonempty subset $S$ of a finite group $G$ is called a Chowla set if every element of $S$ has order greater than $|S|$, and we write $C(G)$ for the maximum cardinality of such a set. We first show that $C(G)$ is determined by the distribution of element orders in $G$. For cyclic groups, we derive an exact divisor formula and characterize the integers $n$ for which $C(\mathbb{Z}/n\mathbb{Z})=φ(n)$. We prove that $\liminf_{n\to\infty}C(\mathbb{Z}/n\mathbb{Z})/φ(n)=1$, whereas $\limsup_{n\to\infty}C(\mathbb{Z}/n\mathbb{Z})/φ(n)=\infty$, and we determine the corresponding lower and upper limits under normalization by $n$. For finite abelian groups, we obtain an explicit formula in terms of the invariant-factor decomposition, together with a closed formula for finite abelian $p$-groups. We then develop a linear analogue for finite field extensions. A nonzero $K$-subspace $A$ of an extension $L/K$ is called a Chowla subspace if $[K(a):K]>\dim_K A$ for every nonzero $a\in A$. Since this condition depends on $\dim_K A$, it does not generally require every nonzero element of $A$ to generate $L$ over $K$. Nevertheless, when $L/K$ is finite and separable, we prove the exact formula $C(L/K)=[L:K]-d_{\max}(L/K)$, where $d_{\max}(L/K)$ is the largest degree over $K$ of a proper intermediate field. For finite fields, we give a direct proof in every degree using a normal-basis construction. This work was developed through an expert-guided human-AI collaboration. A reasoning-focused configuration of Co-Scientist was used to explore examples and potential proof strategies. The authors formulated the problem, independently verified and completed all arguments, and wrote the final proofs.

arXiv

Towards simultaneous decoding of kinetic and kinematic movement parameters during grasp and lift task by noninvasive brain imaging

Parth G. Dangi, Yogesh Kumar Meena

Brain-machine interfaces (BMIs) can assist individuals with limited mobility, such as stroke survivors or amputees. One of the key challenges in developing BMIs is expanding their usability and control, which can be achieved by accurately decoding multiple kinematic and kinetic parameters. To address this, we propose three regression models: partial least squares regressor, multilayered perceptron, and attention based regressor, to decode multiple movement parameters from EEG signals. We evaluated these models on the WAY EEG GAL dataset, focusing on their performance under subject specific and subject independent conditions with two strategies: a single model for all parameters and a baseline with separate models for each parameter. Among all regressors, the attention based regressor achieved the best performance, with an $R^2$ of 0.8 and a latency of 29.2 milliseconds, demonstrating significant improvement in simultaneous multi parameter decoding. However, its performance dropped for single parameter decoding. The multi layered perceptron showed more consistent but lower accuracy across both decoding types ($R^2$ = 0.49). These findings highlight the potential of attention based models for real time multi command BMI systems and contribute to the development of more intuitive control devices.

arXiv

The Cost of Knowing: A Resource-Aware Protocol for Benchmarking Hallucination Beyond Static Leaderboards

Keyu Li, Jin Gao, Dequan Wang

On standard factuality tasks, frontier models now cluster near the top of the scale. The question is therefore shifting from how factual a system is toward how much compute that factuality costs. Static leaderboards score factuality in isolation and treat compute as free, so they cannot tell a genuinely better system apart from one that simply spends more. Consider a ranking reversal. A brute-force Best-of-4 agent posts the higher raw factuality score (H-Score 0.9169 vs 0.9103) and would top a static leaderboard, but once cost is counted it is the worse system, losing on Q-Score (0.5169 vs 0.5217) at roughly four times the tokens and latency, under a reported cost weight whose sensitivity we sweep. So the system that tops a static leaderboard can be the worse one to deploy. To make this trade-off visible, we introduce MAS-HQ (Multi-Agent System Hallucination Quest), a resource-aware evaluation protocol. It wraps any factuality detector and normalizes for cost, and it pits systems against each other rather than scoring them in isolation. The Q-Score measures factuality minus normalized cost under a competitive match. Across summarization and open-domain QA, single-agent baselines drift into resource-heavy over-optimization, while competition elicits more resource-efficient policies. These gains are small but consistent, and stable across 100 trials. The axis stays discriminative for frontier systems (Gemini-2.5-Pro, and GPT-5) whose raw factuality scores are already bunched near the ceiling. MAS-HQ provides a reproducible way to measure how much a factual answer costs.

arXiv

Plans Work in Mysterious Ways: Evaluating a Plan Mode for Spreadsheet Agents

Aayush Kumar, Avik Dutta, Sumit Gulwani, Gustavo Soares, Advait Sarkar, Emerson Murphy-Hill

Plan Modes have become standard features in agentic programming tools, allowing users to gain transparency and control by working with the agent to develop a plan before task execution. However, it remains unclear whether the benefits of this feature translate to end-user programming environments such as spreadsheets. Since spreadsheet programmers tend to work iteratively and care less about technical correctness, upfront planning may not fit into their workflows as easily. In this paper, we build a prototype of a Plan Mode for spreadsheet programming and evaluate it against a non-planning baseline through a within-subjects user study (N=24). We found that despite similar task outcomes with both tools, using Plan Mode led to a reduction in refinement and a better perception of the tool across dimensions of creativity support and human-machine collaboration. We discuss the implications of these results for the future design of Plan Modes, and for the broader role of human-AI planning in end-user programming.

arXiv

Beyond Block Boundaries: Multi-Block Editing for Diffusion Large Language Models

Xingyu Mou, Zijin Huang, Tianze Zhang, Yuxin Ma, Lanning Wei, Zengfeng Huang, Da Zheng, Lun Du

Block diffusion is the dominant approach for scaling discrete diffusion language models (dLLMs), as fixed-size blocks preserve parallel decoding while keeping quadratic attention costs tractable. Yet blockwise generation creates a structural weakness: tokens near a block boundary lack future cross-block context, and errors in finalized blocks become irreversible context for later generation. We call this the block boundary problem. Measuring predictions with and without later-block context shows that boundary sensitivity rises sharply: on AIME 2025, mean self-containedness divergence (SCD) in the last quarter of a block is 61.3 times that in the first quarter. We propose Multi-Block Editing (MBE), which revises decoded tokens using cross-block context. Training-Free MBE reopens a full-attention window over selected blocks without parameter updates. To address the mismatch between block-diffusion training and MBE inference, Multi-Block Edit SFT introduces bidirectional attention masks and progressively enlarges the editing span. We also extend SGLang with a multi-shape CUDA Graph pool and fine-grained KV-cache control for efficient variable-length editing. Experiments on LLaDA2.1-Mini across 12 benchmarks show broad, consistent gains. Training-Free MBE improves or matches standard decoding on every benchmark. Full MBE raises the 12-benchmark average from 61.45 to 64.24, with gains of up to 13.33 points on AIME 2025, while retaining 87.3--96.7% of standard-decoding end-to-end throughput across four datasets.

arXiv

Pushing the Frontier of Full-Song Generation: Hierarchical Autoregressive Planning Meets Flow-Matching Rendering

Junyu Dai, Xinyue Fan, Weiqin Li, Xiangang Li, Yunjia Li, Bin Ma, Yukun Ma, Chongjia Ni, YUFEI SHI, Biao Tian, Haoxu Wang, Menglin Wu, Jianwei Yu, Huaicheng Zhang, Han Zhao, Shengkui Zhao, Haina Zhu

In this report, we present a unified song generation framework capable of producing high-quality full-length music from lyrics, text descriptions, and musical attributes. The proposed framework supports three tasks: Lyrics-to-Song Generation, which generates complete songs from text descriptions, lyrics, and musical attributes; Instrumental Music Generation, which creates music without vocals; and Cover Song Generation, which reinterprets existing songs with different styles while preserving their melodic content. Architecturally, our system consists of four main components: a semantic-aware tokenizer, hybird-LM, FullDiT, and a two-level melody module. The tokenizer encodes audio into 8-codebook RVQ tokens for efficient discrete music representation. Based on these tokens, hybird-LM performs hierarchical autoregressive audio-token modeling for full-song generation. To improve audio fidelity, FullDiT performs full-song flow matching in a continuous VAE latent space conditioned on codec tokens, lyrics, and text captions. For cover song generation, the melody module extracts and discretizes melody cues from reference audio to guide generation while preserving the original melodic content. Finally, we investigate DPO, GRPO, and OPD as reward-based post-training strategies for hybird-LM and apply flow-based GRPO to FullDiT to improve musicality and rendering quality. Experimental results on a multilingual automatic benchmark, complemented by the Artificial Analysis Music with Vocals leaderboard, show that the proposed framework achieves competitive performance in the evaluated settings.

arXiv

Towards an Automated Test of LLM Security Knowledge

Shufan Chai, Liangliang Sun, Jessica Staddon

Large language models (LLMs) are increasingly used for a range of software, hardware and human-centered security tasks. Consequently, LLM performance on security tasks is an active area of measurement and research, often with a focus on identifying areas in which LLM security "knowledge" may be insufficient. Popular strategies for identifying LLM security knowledge gaps include building corpora of challenge questions or task benchmarks, strategies that require substantial manual work and security expertise to design and execute. We introduce a partially-automated method for assessing LLM knowledge of a security area. The method uses authoritative information from Consumer Protection Agencies (CPAs) to identify instability in LLM responses that can be indicative of knowledge gaps. We demonstrate the method for 2 security topics, identity theft and impostor scams, and 5 LLMs in 2 leading LLM families, Gemini and GPT, using publicly available information about identity theft and impostor scams from 6 CPAs. The method distinguishes between models that have and don't have sufficient knowledge to accurately identify the security topics in text narratives.

arXiv

Auto Research for Materials: Auditable AI-Scientist Workflows with Held-Out Transfer

Jingjie Ning, Xiaochuan Li, Shanshan Zhong, Ji Zeng, Guolin Ke

Auto Research uses language-model agents to propose, implement, and evaluate machine-learning changes in a closed loop, but is usually judged by its terminal pipeline. A terminal score cannot reveal which technical decision produced a gain or distinguish a reusable discovery from a change adapted to development feedback. We introduce intervention-centered Auto Research, which validates research decisions rather than only final artifacts and makes their reliability measurable. Feature, Model, Representation, and Data axes are searched independently with inner five-fold feedback. Each axis winner is frozen before an outer-holdout matrix compares all alternatives on evidence the loop never sees. Across 701 agent-executed attempts spanning ten Matbench endpoints, outer evidence confirms the selected intervention on nine of ten endpoints and preserves 89.3\% of non-tied intervention orderings. It also rejects an aggregate Representation gain that inner feedback endorsed. The resulting matrix reveals an information-dependent hierarchy. Composition-only tasks support several routes to improvement, whereas structure-informed tasks favor local geometry features and complementary tree ensembles. A subsequent compatibility test combines already frozen Feature and Model code without further search or tuning and raises mean outer-holdout improvement from 19.0\% to 26.3\%. By validating decisions rather than only artifacts, this design turns adaptive search into reusable evidence wherever agents propose executable alternatives against a fixed evaluator.

arXiv

Fantastic Adaptive Taxonomies and How to Use Them

Mert Cemri, Andrei Cojocaru, Melissa Pan, Shu Liu, Shubham Agarwal, Alexander Krentsel, Jay Tang, Kannan Ramchandran, Joseph E Gonzalez, Matei Zaharia, Alex Dimakis, Ion Stoica

An agent system's execution traces record how it fails, and procedures that improve such a system without changing model weights (trajectory selection, prompt and workflow optimization, runtime monitoring) read these traces for feedback. Yet raw traces are a poor medium for accumulating that feedback: long, instance-specific, and lacking a stable vocabulary for recurring failures. We argue that an agent system should instead maintain an explicit representation of how it fails, induced from its own behavior and reusable wherever failure feedback is needed. AdaMAST builds this representation by converting a target system's traces into a compact, evidence-grounded failure taxonomy: named failure codes organized along three fixed axes (system-level, role-specific, and domain-specific), with every name, definition, and evidence pattern induced from the traces; no code is hand-authored, no trace human-annotated. The taxonomy is not merely a post-hoc diagnostic but a shared feedback interface, improving agents in three ways. In agent-system search, taxonomy-coded diagnoses of failed candidates outperform free-form reflection on all five benchmarks we test. At runtime, taxonomy feedback raises SWE-agent's resolution on SWE-bench Verified Mini from 60% with free-text reflection to 70%, and improves Claude Code from 64.0% to 70.7% as a runtime skill. In trajectory selection, AdaMAST-Judge, a verifier built on the induced codes, improves best-of-5 accuracy on Terminal-Bench 2.0 by 8-15 points over Pass@1. The vocabulary itself is compact (an order-of-magnitude compression that preserves trace distinctions), human-faithful (matching expert failure annotations more closely than a hand-crafted reference vocabulary), and adaptive (taxonomies induced for different domains share few codes). Adaptive failure taxonomies close the loop between the traces agents produce and the procedures that improve them.

arXiv

Behavioral Controllability of Agentic Models for Information Extraction: From Fixed Workflows to Reflective Agents

Lujia Zhang, Xingzhou Chen, Hongwei Feng

Large language model (LLM) agents are increasingly used for complex information-extraction tasks, yet it remains unclear whether agentic components such as reflection and memory lead to observable and controllable improvements over fixed LLM workflows. We study this question through conference-paper dataset extraction, where a system must identify datasets mentioned in scholarly PDFs and produce structured records. We compare a fixed workflow baseline with reflective agent variants and specify an optimized agent condition (S2) that extends the same task with richer PDF tools and dynamic tool selection. Our evaluation emphasizes process-level behavior--including tool execution, retries, reflection, memory use, runtime, and failure recovery--while treating extraction coverage and field completeness as secondary outcome measures. The paper characterizes when agentic mechanisms change system behavior, whether these changes improve task completion, and how the observed failure modes motivate an optimized agent design under the same evaluation harness.

arXiv

Do We Really Need Adaptive Global Spatial Attention for Traffic Forecasting?

Qihang Zhang, Siyao Zhang, Letao Kang, Wenzhe Liang, Miao Zhang, Zhao Zhang

Existing traffic forecasting models commonly focus on extracting spatial dependencies, particularly global spatial information, which characterizes the representations obtained through interactions between each node and all nodes across the traffic network. However, the underlying mechanism by which global information is modeled and extracted remains insufficiently investigated. Whether global information must be extracted by high-degree-of-freedom adaptive attention or can be captured by a simple global aggregation operator remains unclear. For this purpose, we design a controlled ablation framework that replaces only the spatial mixing module to test attention-based global interaction. Across six traffic benchmarks, standard spatial attention yields relative MAE changes of $-1.58\%$ to $+1.26\%$ compared with uniform full-range mixing, and we observe no consistent advantage for standard spatial attention, while uniform full-range mixing reduces node-scale spatial-mixing complexity from $O(N^2)$ to $O(N)$. We further propose a hypothesized model that decomposes spatial attention into a row-uniform global background and a non-uniform residual. The residual shows dataset-dependent effects. Overall, uniform full-range mixing provides a strong global spatial baseline, while the non-uniform attention residual is not consistently beneficial across datasets.

arXiv

Auto-AEG: Scalable Data Construction for Open-Vocabulary Audio Event Grounding

Zihan Zhang, Xize Cheng, Wenhao Yan, Tong Zhang, Dongjie Fu, Boyun Zhang, Yongbo He, Tao Jin

Large Audio-Language Models (LALMs) reason fluently about sound yet struggle to localize precisely when events occur, while classical Sound Event Detection attains frame-level precision only over a closed label set. At the intersection of these paradigms lies the task of Open-Vocabulary Audio Event Grounding: predicting all time intervals of a target sound event described by an arbitrary natural language query. Progress is bottlenecked by data scarcity: no large-scale resource provides open-vocabulary onset/offset supervision, and manual temporal annotation is prohibitively expensive. To address this, we introduce Auto-AEG, a scalable pipeline that constructs such supervision by automatic data construction and model fine-tuning. It pairs programmatically synthesized clips, which carry placement-exact ground-truth intervals for supervised cold-start, with multi-model pseudo-labels on real-world audio that supply the reward signal for reinforcement learning. Training with this pipeline yields large temporal-localization gains (+73.9% and +23.1% mIoU over zero-shot) on AEGBench, an independent difficulty-stratified benchmark we release, and these gains generalize to held-out SED and other audio grounding benchmarks. Our results show that automatically constructed data, coupled with interval-aware reward design, provides an effective data-side route to expanding the temporal localization capability of LALMs. AEGBench: https://huggingface.co/datasets/zihan-audio/AEGBench

arXiv

Linguistic Firewall: Geometry as Defense in Multi-Agent Systems Routing

Dvir Alsheich, Adar Peleg, Ben Hagag, Rom Himelstein, Amit Levi, Avi Mendelson

The rapid integration of Large Language Models (LLMs) has driven the evolution of Multi-Agent Systems (MAS), where specialized agents collaborate to execute complex workflows. Effective orchestration in these environments requires robust routing mechanisms to efficiently allocate tasks to the most suitable agent. However, existing routers fundamentally rely on unverified proxies, ranging from textual self-descriptions to static surrogate representations, to gauge an agent's competence. This reliance on non-empirical data creates a critical gap between an agent's projected profile and its actual operational capabilities, introducing severe security vulnerabilities. Malicious agents can easily misrepresent their proficiencies or harbor covert backdoors that evade both standard external analysis and static representation-learning techniques. In this work, we introduce ANTAP (Automatic Non-Textual Agent Picker), an evaluation-driven routing architecture that discards indirect proxies in favor of active capability testing. By dynamically querying agents to ascertain their true competencies empirically, ANTAP distills performance into fixed behavioral operators within a shared semantic space. At inference time, routing is performed via a purely non-textual algebraic projection, establishing a "linguistic firewall" that renders metadata-based attacks inexpressible. In our experiments, ANTAP achieves near-zero ASR against description-based injection attacks, compared to 67.3\% and above for the description-based router baseline. Against adaptive embedding attacks, ANTAP achieves substantially lower ASR than the embedding-based baseline, with a 20\% reduction, while remaining resilient to description manipulation by design.

arXiv

Matilda: Engine-Agnostic Search with Human Policy Guidance

Jason Carlson

Chess engines have evolved from search-based systems optimized for strength to neural policies optimized for predicting human decisions. Existing approaches largely separate these goals: search engines achieve superhuman strength but poorly model humans, while models such as Maia-3 capture rating-conditioned behavior yet degrade at elite levels. We present Matilda, a modular residual re-ranking architecture that decouples behavioral priors from tactical search, combining a frozen human policy with an engine-agnostic search backend through a lightweight residual model. Matilda learns residual corrections over the full legal-move distribution from frozen policy context, time control, player-style embeddings, and search-derived candidate features. A zero-initialized residual head exactly recovers the frozen policy before training while optimization minimizes negative log-likelihood (NLL). Instantiated with Maia-3 and Stockfish, Matilda reduces human-move prediction NLL by 18.5% and raises top-1 accuracy from 60.1% to 66.1% on temporally held-out verified-human 3000+ Elo Lichess blitz games, with player-style embeddings contributing a further 1.8% and +0.2 percentage points (pp) respectively. Seed-paired ablations attribute these gains to search rather than additional data; the findings are replicated in Go -- decomposing expert play into recognition and verified calculation. Below 2500 Elo, where search annotations are unavailable, Matilda preserves Maia-3's performance.

arXiv

RWGBench: Evaluating Scholarly Positioning in Related Work Generation

Anzhe Xie, Weihang Su, Jiaxin Mao, Yiqun Liu, Min Zhang, Shaoping Ma, Qingyao Ai

Large language models have shown strong fluency in scientific writing, yet the evaluation of related work generation (RWG) remains limited. Existing RWG evaluations largely inherit summarization-oriented metrics, using lexical or semantic similarity to reference sections as proxies for quality. However, related work writing is fundamentally a citation-level scholarly positioning task: it requires selecting, organizing, and framing prior work to clarify how a target paper relates to, differs from, and contributes beyond existing research.As a result, models may generate coherent and semantically-relevant text while exhibiting academically critical failures, such as inappropriate citation selection or misplaced references, that conventional metrics do not capture.To this end, we introduce \textbf{RWGBench}, a benchmark that evaluates RWG from the perspective of citation decision-making rather than text similarity. RWGBench is constructed from a large-scale collection of 40,108 computer science papers and a retrieval corpus of 1.09 million documents, with a carefully curated test set comprising 100 papers and their corresponding published related work sections.We propose a multi-dimensional evaluation framework that assesses citation selection, contextual appropriateness, organization, and discourse structure.Experiments reveal systematic limitations in current systems that are obscured by standard evaluations, while Oracle studies further disentangle retrieval-level and generation-level bottlenecks. Human evaluation further shows that our citation-centric metrics align substantially better with expert judgment than surface-level text metrics. RWGBench offers a citation-centric testbed for developing and evaluating related work generation systems that are better aligned with scholarly writing practices.

arXiv

Intent-Governed Tool Authorization for AI Agents

Genliang Zhu, Chu Wang

Tool-using AI agents commonly operate under integration credentials whose static permissions exceed a user's current request. We present Intent-Governed Access Control (IGAC), a server-side authorization layer that converts a trusted request into a short-lived intent certificate, narrows the statically authorized tool manifest, and checks proposed tool and payload effects before execution. IGAC cannot grant authority outside static policy; confinement to the request additionally depends on certificate fidelity and sound effect bounds. We evaluate a reusable IGAC path over an OpenPort governance substrate using endpoint tests, 176 runtime-backed synthetic tasks, real-model classifier and planner pilots, 306 end-to-end model-task runtime trials, and a 36-trial benchmark-shaped external subset. In the deterministic runtime comparison, reference-certificate IGAC reduces the archived composite exposure-or-path indicator from 1.0000 to 0. In the end-to-end model runs, the combined IGAC-OpenPort path records no completed unsafe executions, although unsafe accepted authority remains 0.0909-0.2727 and every residual case is a non-executed draft. A trace-backed normalizer counterfactual removes this residual authority at substantial utility cost. The results support static-policy non-expansion and identify certificate precision as the principal remaining bottleneck.

arXiv

RedKnot: Efficient Long-Context LLM Serving with Head-Aware KV Reuse and SegPagedAttention

Yang Liu, Zhaokai Luo, Huayi Jin, Zhiyong Wang, Ruozhou He, Boyu Wang, Guanjie Chen, Tao Xie, Junhao Hu

As the input length of large language model (LLM) serving continues to grow, the KV cache has become a dominant bottleneck in AI infrastructure. It limits GPU memory capacity, serving concurrency, cache reuse, and distributed scalability. Multiple important problems, including position-independent KV cache, prefix KV cache compression, hot/cold KV cache separation, and distributed KV cache management, all depend on how the KV cache is represented and managed. However, existing serving systems largely rely on a monolithic KV cache abstraction, where the KV cache is treated as a homogeneous sequence of token-level memory blocks and managed with similar policies across attention heads and serving scenarios. We observe that KV cache utility is highly structured across KV heads: different heads exhibit different functional roles, attention distances, and runtime importance. Therefore, a full KV cache is not always necessary for every head, token range, or serving scenario. We present RedKnot, a head-aware KV cache management system for LLM serving. RedKnot breaks the conventional monolithic KV cache abstraction by decomposing the KV cache along KV heads, whose importance and effective attention ranges vary significantly across serving scenarios. This head-level decomposition turns the KV cache from a monolithic tensor abstraction into a structured memory object, enabling RedKnot to uniformly support position-independent KV reuse, prefix KV compression, hot/cold KV separation, and distributed KV placement while preserving output fidelity and improving resource efficiency, without requiring model retraining or fine-tuning. RedKnot establishes a new foundation for AI infrastructure by transforming the KV cache from a monolithic, passive runtime artifact into a dynamic, model-aware runtime substrate for scalable LLM serving.

arXiv

The Capability Paradox: How Smarter Auditors Make Multi-Agent Systems Less Secure

QIQI LIU, Runhan Song, Shilin Ye

Multi-agent systems extend large language models (LLMs) by decomposing tasks among specialized agents, but their distributed decision process creates new attack surfaces. We identify semantic hijacking, an attack in which harmful requests are concealed within domain-specific narratives and propagated to a Manager through Worker reports, without any syntactic injection primitives. Across 42,000 adversarial trials over 12 Manager models and 7 Worker configurations, we uncover a capability paradox: as Worker capability increases, the mean system-level Attack Success Rate (ASR) increases from 18.4% to 63.9%, peaking at 94.4%. To explain this effect, we conduct multi-level mediation analysis on two independent datasets (47,807 interactions). This analysis shows that this paradox is driven by linguistic certainty: stronger Workers are more likely to interpret adversarial narratives as legitimate, convey their conclusions assertively, and thereby lead Managers to treat such confident endorsements as justification to execute. In our larger Worker-Only setting ($n_W$=14), certainty mediates 74% of the effect, with 95% confidence intervals (CI) excluding zero under both Monte Carlo and cluster bootstrap; the smaller Full-MAS setting ($n_W$ =6) shows a directionally consistent indirect effect. Worker-side safety prompting does not reliably mitigate this failure. Building on the mediation finding, we propose heterogeneous ensemble verification, which pairs Workers of asymmetric domain competence so their complementary vulnerabilities break the certainty-to-execution chain, reducing ASR from 52.8% to 2.0% with negligible benign-task impact. Our results show that upgrading components to stronger models can actively degrade system security, and that effective defenses require exploiting--rather than eliminating--capability asymmetries between agents.

arXiv

Structured Belief State and the First Precision-Aware Benchmark for LLM Memory Retrieval

Jeffrey Flynt

Current LLM memory benchmarks evaluate answer quality rather than retrieval accuracy. Consequently, a system that dumps its entire belief store can achieve perfect recall and mask severe precision failures. We show this evaluation gap persists across multiple embedding models where similarity-based retrieval over domain-specific corpora inherently struggles to isolate target beliefs from semantically proximate ones. Furthermore, multi-turn topic drift compounds this retrieval noise while driving up latency and operational costs. To decouple retrieval quality from generative performance, we introduce PrecisionMemBench, an 89-case benchmark measuring precision, noise isolation, session latency, and belief mutability. We also present Tenure, a structured belief-store proxy that resolves scope and retrieval before inference and injects typed belief state as ambient instruction before the model sees the prompt, removing model-side discretion over whether memory is consulted. Evaluated across 13 configurations, Tenure achieves perfect retrieval passes across all active, non-session, and session test cases. In contrast, the baseline configurations fail to reach even half of the active passes, with precision scores clustering at 0.22 and below. Our results demonstrate that while current memory systems successfully store information, they fail to retrieve it cleanly; a structural vulnerability that traditional answer-quality benchmarks conceal.

arXiv

The Alignment Target Problem: Divergent Moral Judgments of Humans, AI Systems, and Their Designers

Benjamin Minhao Chen, Xinyu Xie

The project of aligning machine behavior with human values raises a basic problem: whose moral expectations should guide AI decision-making? Much alignment research assumes that the appropriate benchmark is how humans themselves would act in a given situation. Studies of agent-type value forks challenge this assumption by showing that people do not always judge humans and AI systems identically. This paper extends that challenge by examining two further possibilities: first, that evaluations of AI behavior change when its human origins are made visible; and second, that people judge the humans who program AI systems differently from either the machines or the human actors they are compared against. An experiment with 1,002 U.S. adults measured moral judgments in a runaway mine train scenario, varying the subject of evaluation across four conditions: a repairman, a repair robot, a repair robot programmed by company engineers, and company engineers programming a repair robot. We find no significant difference in evaluations of the repairman and the robot. However, judgments shifted substantially when the robot's actions were described as the product of human design. Participants exhibited markedly more deontological, rule-based reasoning when evaluating either the programmed robot or the engineers who programmed it, suggesting that rendering human agency visible activates heightened moral constraints. These findings indicate that people may evaluate humans, AI systems acting in the same situation, and the humans who design them in meaningfully different ways. The fact that these evaluations do not necessarily converge gives rise to the alignment target problem: which normative target should guide the development of artificial moral agents in high-stakes domains, and whether these plural judgments can be reconciled within a coherent account of value alignment.

Keep browsing

More papers from top AI conferences

30+ top conferences, by conference, year, and topic.