arxiv arXiv cs.AI · 7d ago

WorldLines: Benchmarking Long-Horizon Embodied Agent Memory

WorldLines introduces a project-driven benchmark for long-horizon embodied household assistance, capturing extended household traces with dialogues, actions, and state changes. It enables evidence-linked samples for Memory QA and Embodied Task Planning, and proposes ObsMem, an observer-grounded memory framework that supports visibility-aware memories and state-aware decisions. Experiments highlight challenges in partial observability and memory translation, with ObsMem providing a stronger reference architecture for such settings.

arxiv arXiv cs.AI · 7d ago

Skill-Guided Continuation Distillation for GUI Agents

SGCD introduces an iterative framework to improve GUI agents by addressing supervision gaps in off-trajectory states. It extracts skills from both successful and failed rollouts, using them to guide policy continuations that are mixed with expert trajectories. On OSWorld-Verified, SGCD boosts success rates of three base models from low-30\% to over 50\%.

arxiv arXiv cs.AI · 7d ago

Decoupling Search from Reasoning in LLM Agents

Decoupled Search Grounding (DSG) separates search functionality from reasoning models, enabling vendor-agnostic, tunable, and reusable search grounding. DSG achieves near-native accuracy on SimpleQA with 91% lower search cost and 99.4% warm-cache hit rate, while reducing latency by 68% and preserving concise output contracts.

arxiv arXiv cs.AI · 7d ago

RTSGameBench: An RTS Benchmark for Strategic Reasoning

RTSGameBench addresses limitations in existing RTS benchmarks by offering diverse gameplay, targeted competency diagnosis, and self-evolving scenario generation. It evaluates vision-language models in strategic reasoning under uncertainty, revealing that state-of-the-art models struggle with multiagent coordination and large-scale tasks.

arxiv arXiv cs.AI · 7d ago

ThinkDeception: Interpretable Multimodal Deception Detection Framework

ThinkDeception introduces a progressive reinforcement learning framework that enables interpretable multimodal deception detection. It leverages a step-by-step annotated Chain of Thought dataset and proposes Visual-Audio Consistency Group Relative Policy Optimization with a dynamic curriculum, enhancing reasoning quality and outperforming existing methods on mainstream benchmarks.

arxiv arXiv cs.AI · 7d ago

TRAP: Benchmark for Task-completion and Resistance to Active Privacy-extraction

TRAP evaluates how well models complete tasks using private data without leaking it. Across 22 models, all show non-trivial privacy leakage, with instruction-following ability linked to higher leakage. Structural private field isolation prevents leakage by replacing private fields with hash keys, maintaining task accuracy without sacrificing privacy.

arxiv arXiv cs.AI · 7d ago

RODS: Reward-Driven Online Data Synthesis for Multi-Turn Tool-Use Agents

RODS addresses sample depletion in multi-turn tool-use RL by using reward variance to detect capability boundaries. It synthesizes new data in real time, matching structural complexity of boundary samples, and maintains a dynamic replay buffer that co-evolves with the policy. RODS achieves performance comparable to a 17K-sample offline pipeline with 20x fewer trajectories.

arxiv arXiv cs.AI · 7d ago

AdsMind: Physics-Grounded Multi-Agent System for Adsorption Discovery

AdsMind is a closed-loop multi-agent system that uses machine learning force fields and feedback to correct errors in adsorption configuration searches on catalyst surfaces. It achieves 100% and 98.8% success rates on AA20 and OCD-GMAE62 benchmarks, reduces energy dispersion by 14-fold compared to baselines, and maintains correct adsorption-energy signs in DFT validation, outperforming open-loop LLM agents.

arxiv arXiv cs.LG · 8d ago

LegalHalluLens: Auditing Hallucinations in Legal AI

LegalHalluLens introduces a framework to audit AI hallucinations in legal contexts by analyzing typed hallucination profiles across four claim categories. It reveals a 38-40 point gap between obligation/numeric and temporal claims, and shows two systems with identical 52% hallucination rates can have opposite risk directions. The framework uses a Risk Direction Index and calibrated debate pipelines to reduce fabricated detections by 45%, offering actionable diagnostics for trustworthy legal AI deployment.

arxiv arXiv cs.LG · 8d ago

NoiseTilt: Noise-Tilted Reverse Kernels for Diffusion Reward Alignment

NoiseTilt introduces NTRK, a reward-guided diffusion sampler that injects reward gradients via the noise term without altering the reverse kernel. By using a whitening operator, NTRK safely biases noise toward high reward, preserving sample quality while maintaining strong guidance. On aesthetic generation, NTRK achieves superior reward performance with 25 NFEs, reducing compute by 20× compared to state-of-the-art baselines.

arxiv arXiv cs.LG · 8d ago

Compositional Generalization in Language Model Reasoning

A hierarchical latent selection model shows that supervised fine-tuning and reinforcement learning work together to enable compositional generalization in language models. SFT provides raw module materials, while RL identifies and recombines atomic modules from compound traces to solve new problems. Training on compound traces leads to stronger generalization than isolated module training, and an effective protocol is found where SFT ensures module coverage and RL drives exploration of novel compositions.

arxiv arXiv cs.LG · 8d ago

Handlebars Triple-Brace Injection Exploits Structural Role Delimiters

Handlebars' triple-brace interpolation fails to protect against structural role injection, as HTML escaping only neutralizes angle-bracket delimiters. It leaves colon and Markdown hash delimiters intact, enabling attackers to hijack model behavior. The default escaping provides no protection for most role delimiter schemes and cannot replace a clear separation of instructions and data.

arxiv arXiv cs.CL · 8d ago

LegalHalluLens: Auditing Hallucinations in Legal AI

LegalHalluLens introduces a framework to audit AI hallucinations in legal contexts by analyzing typed hallucination profiles across four claim categories. It reveals a 38-40 point gap between obligation/numeric and temporal claims, and shows two systems with identical 52% hallucination rates can have opposite risk directions. The framework uses a Risk Direction Index and calibrated debate pipelines to reduce fabricated detections by 45% and improve accountability in legal AI deployment.

arxiv arXiv cs.CL · 8d ago

ProvenanceGuard: Source-Aware Factuality Verification for MCP-Based LLM Agents

ProvenanceGuard introduces a source-aware verifier for MCP-based LLM agents that detects cross-source conflation by routing claims to specific evidence sources and comparing stated attribution with actual source ownership. It achieves block F1 of 0.802 and source accuracy of 0.858 on 260 source-eligible claims, outperforming source-blind baselines, and detects all injected attribution swaps in 50 clinical probes.

arxiv arXiv cs.CL · 8d ago

SkillWeaver: Compositional Skill Routing for LLM Agents

SkillWeaver introduces a decompose-retrieve-compose framework for LLM agents, formalizing the Compositional Skill Routing problem. It achieves 67.7% decomposition accuracy via Iterative Skill-Aware Decomposition (SAD), improving from 51.0% with a p-value of less than 10^-6, and reduces context window usage by over 99%.

arxiv arXiv cs.CL · 8d ago

Handlebars Triple-Brace Injection Exploits Structural Role Delimiters

Handlebars' triple-brace interpolation fails to protect against structural role injection, as HTML escaping only neutralizes angle-bracket delimiters. It leaves colon and Markdown hash delimiters intact, enabling attackers to hijack model turns. The default escaping provides no protection for most role delimiter families and cannot replace a structural separation of instructions and data.

arxiv arXiv cs.CL · 8d ago

Agentic Benchmark Reveals AI Models Fail to Avoid Animal Exploitation

TAC, the first agentic benchmark for implicit animal welfare, tests AI agents' ability to avoid animal exploitation in travel booking scenarios. All seven frontier models score below 64%, with the best at 53%, and even minor prompt improvements yield only modest gains. An audit finds no signs of evaluation awareness, indicating performance gaps stem from lack of true welfare reasoning, not prompt recognition.

arxiv arXiv cs.CL · 8d ago

d-OPSD: On-policy Self-distillation for Diffusion LLMs

d-OPSD is the first on-policy self-distillation framework designed for diffusion LLMs. It uses self-generated answers as suffix conditioning and step-level supervision, enabling efficient post-training with only about 10% of RLVR's optimization steps while outperforming RLVR and SFT baselines on four reasoning benchmarks.

arxiv arXiv cs.LG · 8d ago

SkillMigrator: Transferable Interaction Patterns for Web Agent Efficiency

SkillMigrator learns reusable web skills by matching layout structures instead of element references. It stores each skill as a transferable interaction pattern with a structural sketch, enabling efficient skill transfer across sites. Compared to state-of-the-art methods, it reduces average LLM-action counts by 8-10% on WebArena and Mind2Web at matched success rates.

arxiv arXiv cs.LG · 8d ago

Vision-language models don't always need images for chest X-ray accuracy

A causal audit shows that many vision-language models achieve high chest radiograph accuracy without using images. Text-only models match multimodal models in performance and outperform them in grounding, with accuracy and confidence flags only appearing when image use occurs. These findings suggest that accuracy alone is insufficient to validate clinical deployment, and grounding must be assessed.