All papers

Develops a post-training framework that trains computer-use agents for video games using synthetic interactive experiences and an expanded action space.

The authors introduce FDE-Gym, a benchmark of 95 customer delivery tasks evaluating AI agents on ambiguous requirements and live software environments.

The paper introduces Agent-Reflex, a training framework designed to teach language agents how to reflect and recover from noisy environment feedback.

The paper formalizes a framework and theoretical generalization bounds for evolving an agent's software harness without updating model weights.

The paper provides an open platform and environment suite for evaluating and training GUI agents with online reinforcement learning.

The authors propose a two-level recursive self-improvement framework to evolve and refine execution harnesses for open-world agents.

The paper presents a dataset and framework with dense progressive rewards to train language agents on long-horizon tasks.

The paper introduces a benchmark evaluating how well language models engineer isolated execution harnesses for coding agents using specification documents.

This paper presents a framework that converts long-term real user behavior records into supervisory tasks to train personalization-focused language models.

This paper proposes an on-policy distillation method that aligns student agents using semantic actions rather than token-level supervision.

Introduces ProgramGym, a framework providing verifiable sandbox environments and black-box test feedback for training long-horizon program reproduction agents.

Introduces a framework and multimodal dataset of human desktop demonstrations to induce reusable skills for computer-use agents.

The paper introduces a benchmark measuring how software interface versions affect the task performance of desktop language model agents.

The paper presents a benchmark of 600 tasks to evaluate how well multi-agent vision-language systems handle conflicting information during computer use.

The paper proposes an edge-cloud collaboration framework that selectively delegates complex agent steps from local models to cloud models.

Develops small multimodal models that output structured choices using masked language modeling to reduce inference latency in agent tasks.

This paper introduces a benchmark with expert-written rubrics to evaluate deep research agents on real-world life and productivity tasks.

Introduces a framework where a small model drafts tool-use trajectories and a larger model verifies them in single prefill passes.

The paper presents a benchmark and harness to delegate routine tool choices from large language models to cheaper decision models.

The paper introduces a benchmark to measure how tool-calling agents perform when interacting with emotionally varied, impatient users.

The paper presents a benchmark evaluating whether conversational agents agree on verifiable plans that satisfy both parties' private conditions.

The paper proposes a training-free framework that uses uncertainty signals to escalate computer-use tasks from weak models to strong models.

This paper introduces UnCoBench to benchmark how language model agents handle unhelpful, incomplete, or missing communications from partner agents.

The paper proposes a vision-language-action model where action execution avoids inference latency by decoupling from explicit reasoning output at runtime.

The authors identify a derivation gap where LLM judges reject correct tool arguments lacking direct textual references, and propose a dedicated reference model.

Presents a framework alternating multi-agent sessions across context resets to overcome reasoning degradation in long-horizon machine learning engineering.

Presents a structural scoring framework to select high-quality coding agent trajectories for supervised fine-tuning based on interaction efficiency.

The paper presents PrivacyGuard, a framework using source-linked sanitization and local references to prevent personal data leakage in cloud GUI agents.

Introduces behavioral self-rewarding to refine reinforcement learning feedback for software engineering agents by comparing execution trajectories.

This paper introduces a 240-task benchmark to evaluate recursive self-improvement in agents across diverse real-world office workflows.

Introduces a benchmark evaluating how misleading visualizations and charts affect the multi-step execution and decisions of web agents.

Develops a reinforcement-learned meta-controller for execution agents to decide when to consult stronger advisor language models cost-effectively.

Analyzes archived financial forecasts from multi-model systems to distinguish consensus reliability from baseline underlying asset volatility.

Analyzes how random hardware bit faults propagate and manifest across multi-step LLM agent workflow executions.

The study reports that natural code transformations confuse vulnerability analysis agents, proposing functional changes as a potential defensive strategy.

The authors run controlled experiments showing how parameter schema differences in metadata alter language model tool selection.

The paper demonstrates a hybrid architecture where cloud-generated agent code delegates localized private-data queries to edge models.

The paper introduces a maturity model that grades artificial intelligence agents across multiple capability levels, evidence stages, and safety gates.

The authors analyze tool graph memory in multi-turn agents and find that stored transitions provide little benefit over baselines.

This paper benchmarks language model agents against skilled humans on binary software reverse engineering tasks using interactive tools.

This paper presents a reinforcement learning framework to train language model agents across tools and servers in the Model Context Protocol ecosystem.

This paper evaluates open-weight language models against closed frontier models on a cybersecurity detection benchmark using an interactive telemetry environment.

The paper introduces a diagnostic protocol to evaluate how speech recognition errors affect structured tool calling in voice agents.

The paper studies compute allocation across multi-agent coding teams to identify Pareto-optimal trade-offs between cost and performance.

The paper tests whether role-playing language models leak private or unobserved character information across simulated interactive scenes.

The paper proposes an agent framework that discovers lightweight proxy evaluators to bypass costly benchmark evaluations during autonomous engineering discovery.

The paper introduces BRIDGE, a runtime intervention framework that detects and repairs broken information handoffs across dependency graphs in multi-agent workflows.

Presents a randomized experiment framework to test and curate self-written agent playbooks, preventing task failure caused by unverified lessons.

Shows linear prompt-injection monitors overfit to conversation features and reduces their false alarms by adding matched harmless negative examples.

This paper introduces a system to ensure tool-using language model agents comply with natural language company procedural manuals.

This paper introduces RelLedger, an auditing framework to trace how relational histories drive reporting choices and actions in multi-agent systems.

This paper presents CacheFuse, a training-free technique enabling heterogeneous language models to communicate directly by sharing key-value cache states.

The paper introduces MemForm to benchmark adversarial memory formation poisoning across five input interfaces in language model agents.

The paper introduces a benchmark derived from real customer support dialogues to evaluate language models executing bounded, tool-using skills.

Measures tool-agent performance failures under controlled harness limits such as completion caps and tool menu visibility.

Combines dual-encoder semantic retrieval with a contextual bandit to adaptively select agent tools using execution reliability feedback.

Presents an automated red-teaming framework that searches for harmful compositional tool-use paths across multi-service environments.

Introduces DREAM, an agent-intrinsic defense that triggers functional forgetting of poisoned memory entries through implicit reactivation and topological diagnosis.

Introduces UPTURN, a framework that trains a small classifier to inject redirection prompts as new user turns during reasoning failures.

Develops a framework for evolving deterministic formal policies to serve as runtime guardrails for large language model agents.