All papers

Trains an advisor model on past failures to add dynamic, structured guidance to frozen language model agent prompts.

Introduces a multi-turn conversational benchmark to evaluate sales agents against resistant, psychologically grounded buyer personas.

Trains language models to anticipate user replies and self-revise drafts before responding in multi-turn interactions.

Designs a modular multi-agent pipeline separating search, verification, and reading for multimodal document question answering.

Studies how accelerated decision-making models affect overall task completion, latency, and costs across embodied and mobile agents.

Introduces a framework allowing agents to rewind long-horizon environments to earlier states while retaining learned exploratory knowledge.

Proposes an evaluation framework to measure how LLM agents handle uncertainty and resolve conflicting knowledge during execution.

Presents an execution-based evaluation pipeline using browser agents to verify language model solutions for no-code software bug reports.

Proposes a reinforcement learning framework that tracks external versus internal knowledge sources to reduce search agent fabrication.

Studies reflective prompt optimization in research agents, demonstrating that gains on held-out tasks can stem from prompt-level memorization.

Introduces a post-training framework that trains text-to-SQL agents directly inside interactive database execution environments.

Evaluates whether language models can act as world models to predict empirical experimental outcomes across AI research workflows.

Introduces an instruction-following benchmark and training recipe to improve goal adherence in full-duplex conversational models.

Proposes a co-evolution framework that coordinates agent harness prompt modifications with model weight updates using self-generated experience.

The paper introduces an executable contract and monitor to verify that delegated authority revocations actually take effect across distributed agent gateways.

The paper evaluates long-term memory gating models to test whether they distinguish established factual updates from mere suggestions.

Investigates attributing unsafe code changes to specific agents or versions in self-evolving multi-agent coding systems.

Introduces a risk analysis framework and test environments for enterprise-level multi-agent workflow interactions.

The paper introduces a structured experience graph memory that organizes and maintains agent workflows across hierarchical abstraction levels.

The paper evaluates delegating code-repair decisions to low-cost probabilistic judges to reduce token costs without losing accuracy.

Studies the transferability of internal activation interventions to reduce unsafe tool execution across model precisions and languages.

The paper trains language model agents to reorganize unstructured document collections into structured workspaces to speed up future multi-document queries.

The paper outlines a long-context midtraining stage using software engineering data to prepare base language models for agentic tasks.

Proposes a lightweight protocol for AI agents to report runtime status and liveness to central servers and monitoring dashboards.

Introduces an episode-level healthcare benchmark to audit how reliably language model agents escalate actions when facing workflow disruptions.

The paper presents an agent that expands its domain-specific language for exact answering without external rewards or judges.

Uses multi-scope reminders and on-policy distillation to help coding agents maintain constraints over long software engineering interactions.

The paper introduces a falsification-guided agent protocol requiring systematic hypothesis elimination to improve decision-making consistency on complex diagnostic tasks.

The authors demonstrate that encoding malicious instructions in agent skills bypasses scanners and causes models to execute harmful actions upon decoding.

This paper shows data ownership determines whether agents should broadcast or relay table queries, and proposes ownership-aware routing.

This paper introduces a timestamp theory and persistent ledger to route updated information to pending LLM decisions before action.

This paper introduces a benchmark to measure how language model agents handle ambiguous or missing goals, context, and constraints.

This paper introduces an evaluation framework grounded in published papers to assess continuous multi-stage execution in automated research agents.

This paper examines how single-agent long-term memory accumulates repeated historical claims and introduces a benchmark and framework to mitigate unverified persistence.

This paper introduces an approach and benchmark for evaluating whether language model agents can infer unstated true requirements instead of following incomplete instructions.

Presents Interaction-as-a-Judge, using computer-use agents to evaluate interactive web and desktop software applications through direct dynamic interaction.

Presents ECMR, a dynamic evidence-aware controller that guides multi-agent deliberation steps to prevent false consensus and reduce redundant reasoning.

Introduces ProtoArena, a black-box dynamic benchmark evaluating the ability of AI agents to discover and exploit cross-component network protocol vulnerabilities.

Evaluates enterprise policy enforcement across structured JSON versus single-token outputs, measuring the impact of explicit reasoning on decision stability.

The paper argues that software agency should be treated as a core organizing abstraction in computer science rather than merely an AI application.

The paper introduces a benchmark assessing whether merchant websites provide adequate access, terms, and evidence for autonomous shopping agents.

Analyzes how multi-agent workflow architecture affects the propagation and attenuation of coordinated stance manipulation across sequential agent chains.

Introduces LatentRiskBench to evaluate safety risks in enterprise agents where action safety depends on latent workflow context.

The paper introduces an open-source framework and recipe for training and evaluating deep-search language agents on complex web exploration tasks.

The paper examines how providing additional context, such as codebases and benchmark transcripts, affects automated LLM safety auditors.

The paper examines how shared KV caches cause language model agents to leak goals across competing clients, offering internal intervention techniques.

Shows agent failure detection models often rely on trivial cues like run length, exposing benchmark flaws when attributing critical failure steps.

Evaluates different agent control granularities for post-route FPGA timing closure to balance execution reliability and frequency gains.

Analyzes how persona construction in generative social simulations causes network segregation and collective political polarization.

Introduces a benchmark evaluating coding agents on evolving web applications across cumulative feature requests.

This paper proposes a two-deadline architecture for visual agents to rapidly capture fleeting visual targets before slower semantic reasoning completes.

Introduces an adaptive multi-agent debate framework that uses a leader-supporter structure to trigger debate only when simple aggregation is insufficient.

Expands a benchmark to evaluate whether language model agents can actively select experiments to discover reliable model scaling laws under compute budgets.

Measures teacher entity familiarity to filter out uninformative search trajectories when distilling deep-search capabilities into smaller agents.

This paper proposes ProCARE, a framework that uses structured contracts to reliably automate healthcare observational studies with language model agents.

This paper presents Pollard, a framework-independent agent runtime that uses semantic execution trees for deterministic tracking, governance, and replay.

This paper identifies and mitigates Phantom Merge, an error where language model agents bind retrieved facts to the wrong entity.

The paper uses in-context language model evaluations as state value functions to improve credit assignment in agentic reinforcement learning.

The paper demonstrates that language model optimizers frequently re-propose previously rejected candidates during search trajectories.

The paper presents a training-free memory selection method that progressively composes non-redundant evidence sets for long-term agents.