ZOTPAPER

Codetta protocol enables high-capacity, keyless, undetectable steganography for LLM agents

Qi Pang, Virginia Smith, and Wenting Zheng propose Codetta, a steganographic protocol that allows independently deployed LLM agents to communicate covertly without a pre-shared key. The protocol achieves high capacity while maintaining provable undetectability. On three agent workloads, Codetta transmits up to 94 times more bits per visible token than the prior state-of-the-art asymmetric protocol.

3 Oct 2026

Evolved prover interfaces make theorem-solving agents faster and cheaper

Viennot, Baudart, and Lelarge treat the interface between an AI agent and a proof assistant as an optimization target. Their evolutionary process built rocq-mcp-evolve, a set of agent-facing tools that outperformed both a compiler-only interface and an established Rocq server across four language models.

Memory Tests Must Separate What Agents Keep From What They Retrieve

Huang evaluates retention, which determines what remains in bounded memory, separately from selection, which chooses stored information after a query arrives. Across 300 matched episodes, query-aware selection improved required-fact recall by 15.5 percentage points when access was held constant, while a confounded comparison suggested a much larger 68.7-point gain.

Adaptive agent teams improve consistency in long-form subtitle translation

Jin et al. developed SMART, a multi-agent subtitle translation system that adapts its workflow to each series without changing the underlying language models. It achieved the lowest translation-error penalty across all 15 directions in the authors’ new benchmark, improving by 6.9% on average over the strongest competing agent system.

Evidence-driven harness edits improve frozen GUI agents across benchmarks

Yang et al. introduce an automated method for improving the runtime harness around a GUI agent without changing the underlying model. Across six backbones on OSWorld-Verified, the authors report consistent held-out improvements, including a 12.33-point gain for Qwen3-VL-32B-Instruct. A harness optimized on one benchmark also transferred to WindowsAgentArena, where it improved GPT-5 by 13.87 percentage points.

Agent performance depends on matching models, harnesses and tasks

Li and colleagues evaluate language models together with the software harnesses that expose tools, manage context and return errors. Across 66 configurations and 6,204 scored trajectories, they find no consistently best model or harness: rankings often reverse when either the harness or task changes, while higher spending and vendor-native pairings do not reliably improve results.

Reinforcement learning helps coding agents finish tasks across benchmarks

Mehta, Ritchie and Chen post-trained Kimi K2.7 Code using reinforcement learning with verifiable rewards, without a supervised warm-up. The resulting model achieved higher pass@1 on every evaluated benchmark, including later-released task sets, unfamiliar agent harnesses and multi-hour software projects.

Coding agents can improve by evolving their own instruction rules

Jiang et al. developed RuleEvolve, a framework that automatically refines the instruction files used to guide AI coding agents. Across two agent frameworks, four backbone language models, and three coding benchmarks, the evolved rules matched or improved functional correctness while often producing shorter code and using fewer generation resources than manual rules and prompt-optimization baselines.

Adaptive training curricula produce stronger optimized agent harnesses

Park et al. treat scenario selection as part of agent harness optimization rather than as a fixed input to it. Their system, ActiveSaddler, adaptively revisits unresolved failure types while searching for new weaknesses, raising test Pass@1 by 4.4 percentage points on GAIA2 and 7.5 points on Terminal-Bench 2.0 relative to the same optimizer using a predetermined scenario order.

AI agent teams coordinate worse than a single shared agent

Paliskara, Namchittai and Lampinen test whether AI agents can coordinate when they represent different users competing over a shared resource. Across five frontier models and 77 scenarios, multi-agent teams underperformed a single coordinating agent in every environment, even when teammates could communicate.

Real-time GUI feedback improves training for computer-use agents

Du et al. introduce ComputerSD, a training method that learns from feedback generated after each action a computer-use agent takes. On OSWorld-Verified, it improved task success over outcome-only reinforcement learning by 1.9 percentage points for Qwen3-VL-8B-Thinking and 4.1 points for EvoCUA-8B.

Agentic verification improves outputs on long-horizon workspace tasks

Zhang et al. test whether a fixed language model can verify its own long-horizon work without reference answers, grading rubrics or a stronger judge. Across five workspace benchmarks and two frontier models, their harness outperformed the evaluated selection baselines, while evidence-backed revision improved average scores over a single rollout by more than six points.

Revision-aware agent graphs reuse valid work while avoiding stale answers

Luo and colleagues study dynamic task routing, where an agent must identify which version of a document applied at a query time and then answer the associated problem. Across 373,428 queries derived from six established benchmarks, their Revision-Aware Independent Agent Graph achieved higher joint routing-and-answer accuracy than comparison methods while using substantially fewer model calls.

Dependency graphs pinpoint decisive errors in long agent execution traces

Deng et al. present DeFA, a framework for diagnosing failures whose causes and visible consequences occur many steps apart. Rather than judging each step independently, DeFA maps dependencies across an execution and traces how suspicious events influence later actions. It outperformed seven baselines on text benchmarks and produced feedback that improved a downstream skill-learning pipeline.

Visual memory harness helps multimodal agents solve long interactive tasks

Han et al. introduce VISTA, a model-agnostic harness designed to extend multimodal agents beyond their immediate visual context. Rather than compressing each observation once and relying on that representation throughout a task, VISTA stores past images in their original form and lets the model actively retrieve and arrange them as reasoning proceeds. With Claude Opus 5.0, this approach completed all 25 public ARC-AGI-3 games and achieved the benchmark’s maximum Relative Human Action Efficiency score.

Randomized retrieval reveals which AI memories actually help

Behnam and Wang identify a blind spot in utility-based memory management: a stored memory cannot be evaluated if the retrieval system never places it in the model’s context. Their Causal Memory Policy reserves context slots for randomized memory exposure, improving the ability to distinguish required from non-required memories, while also showing that query-specific utility does not reliably predict future value.

Training Forecasters to Search Improves Calibration and Reduces Tool Use

Afifi et al. trained an agent to research and forecast resolved real-world events, rather than supplying it with evidence collected in advance. On 265 held-out Polymarket questions, the resulting policy improved calibration by 30 to 40 percent, searched less often, and narrowly outscored four tested frontier models under the same evidence-based forecasting harness.

GUI agents improve more when experience is routed by component

Wu, Ding and Huang test whether GUI agents should learn from prior trajectories through fine-tuning or by retrieving notes into the prompt. Rather than choosing one destination for an entire trajectory, they split experience into four component types and route each using measurable properties, improving held-out success over either destination alone.

Active audio-visual tool use improves reasoning across long recordings

Wang and colleagues turn an omni-modal language model into an agent that chooses when to listen, when to look, and which temporal window to inspect. Trained on 170,000 synthetic reasoning trajectories and refined with reinforcement learning, the system reportedly improves performance across audio-visual benchmarks while reducing reliance on single-modality shortcuts.

Reusable reasoning choices make multimodal search agents substantially faster

Zhu et al. introduce Selection-based Structured Reasoning (SSR), which asks a multimodal search agent to choose among reusable natural-language reasoning templates rather than generate a new rationale token by token before every action. Across seven benchmarks with 2B and 4B parameter models, the authors report performance competitive with similarly sized search agents, alongside more than 90% lower per-turn reasoning latency and 28% to 54% lower total model inference latency per question.

Explicit belief states help agents stay coherent over long tasks

Luo and colleagues introduce PoS, an inference-time framework that replaces reliance on accumulated interaction history with an explicit, continually maintained belief state. Across four execution and diagnosis benchmarks and three language-model backbones, the authors report that PoS achieved the best overall performance for every benchmark and backbone combination.

Event graphs improve LLM stock forecasts across two major markets

Liu, Liu and Hu tested an LLM-agent framework that links company news into issuer-specific event sequences while restricting every forecast to information available at that time. Across daily Nasdaq-100 and Hang Seng Index panels from 2024 to 2026, their method outperformed the evaluated LLM agents and momentum baselines on predictive and portfolio metrics.

Shared scheduler enables realistic training of continual-learning agents

Jung et al. introduce SCLATE, an execution substrate for evaluating and post-training agents across sessions, background jobs and memory consolidation events. Across seven ported benchmarks, the system enables controlled comparisons of unmodified agent configurations, while post-training Qwen3.5-4B improved both software-engineering performance and held-out continual-learning accuracy.

Local clinical agents learn to consult remote models without identifying patients

Wang and colleagues developed a multi-agent clinical system in which a local model selectively consults remote specialists while controlling what patient information leaves the institution. On an emergency-department benchmark derived from MIMIC-IV-ED, the system improved task accuracy while sharply reducing both direct disclosure and re-identification from combinations of otherwise indirect clues.

Agents can budget external checks without sacrificing task quality

Di and colleagues formulate external-state checking for stored intentions as a resource-allocation problem under a hard episode budget. Across two benchmarks and three model backbones, their static policy preserved nearly all unconstrained task quality while making substantially fewer observations, while a future-aware policy performed better when observation capacity was scarce.

Evolved agent harnesses cut long-video analysis costs while improving accuracy

Liang, Wu and Dong automate the design of agent harnesses, programs that control which video frames a frozen vision-language model observes and when it answers. Their system combines Monte Carlo tree search, staged candidate evaluation and budget-specific routing, outperforming the strongest hand-crafted video agent by as much as 11.2 points across long-video question-answering benchmarks.

MILO co-evolves agent harnesses and the strategies that discover them

Jana et al. introduce MILO, a framework that automatically rewrites complete agent harnesses while also adapting how it searches for better designs. Using both a frontier model and an open-weight model, the authors report larger performance gains than prior search methods across software engineering, terminal-use and scientific-paper replication tasks.

Models Learn to Allocate Context and Scale Inference More Effectively

Li and colleagues study contextual reasoning, the ability to allocate multiple context windows and decide what information should pass between them during inference. Their Hermes harness gives models varying degrees of control over those decisions, while Hermes-Learn uses supervised fine-tuning followed by reinforcement learning to teach smaller models how to use that control. The authors report that training produces adaptive strategies and performance gains that persist across benchmarks, models, and inference budgets.

Agents learn to reset noisy context during long web searches

Ma et al. present a search harness that lets an agent decide when to preserve useful evidence, discard a polluted context, or pursue another search path. Their 35-billion-parameter model scores 72.83 on BrowseComp and improves over its base model across several multilingual and domain-specific search benchmarks.

Managing research ideas helps autonomous agents search faster

Choi et al. introduce Agentic Idea Manager, a framework that treats research ideas, rather than executable solutions, as the primary units of search. Across 10 AutoLab tasks, AIM outperformed the strongest reported baseline in two task groups and matched baseline performance up to 3.1 times faster.

Reusable harness components let agents adapt workflows to each task

Kuang and colleagues argue that a single agent harness is poorly suited to heterogeneous tasks because mechanisms such as verification, context management and termination logic can help in one setting but distract or slow an agent in another. Their STITCH framework composes task-specific harnesses from reusable primitives, improving reported task success by up to 12 percentage points over fixed harnesses while adding 2.7% test-time overhead.

Agents improve collaboration by learning teams and evidence-gated communication

Huang and colleagues present CollabFlow, a multi-agent system that learns how to assemble agents, connect them and regulate their communication using results from previous tasks. Across 12 datasets, the authors report that this recursive training process outperforms all tested baselines and continues improving over successive rounds.

EvoSteer Repairs Multi-Agent Workflows While They Are Still Running

Zhang et al. present EvoSteer, an orchestration method that can modify a team of tool-using agents before its current task has finished. Across 12 datasets spanning question answering, mathematics, code generation and interactive decision making, the authors report significant improvements over existing orchestration baselines, although the provided material does not include effect sizes.

RefCon improves agent memory by refining and contrasting past experience

Prathama et al. introduce RefCon, a memory-extraction method for agents operating across long sequences of interactions. By combining repeated self-refinement with comparisons among parallel candidate memories, it improves several context-evolving agent frameworks and continues benefiting from additional trajectories where diversity-only sampling begins to saturate.

Terminal agents verify often but miss and mishandle many errors

Luo et al. examined whether terminal agents can accurately assess and correct their own command-line solutions. Across ten agents on TerminalBench2.1, verification was nearly universal, but agents detected only 61.43 percent of incorrect candidates and repaired fewer than half of the errors they detected.

Explicit diversity training gives LLM agents distinct successful strategies

Fu et al. introduce Trajectory-guided Joint Policy Optimization, or TJPO, a reinforcement learning method that trains one language-model policy to produce behaviorally distinct solutions. On Sokoban and ALFWorld, the method increased diversity along user-specified dimensions, while retaining competitive task performance and improving robustness in targeted tests.

Selective self-distillation helps LLM judges generalize on subjective evaluations

Hong et al. train LLM judges using preference rationales and rubrics rather than relying only on rewards for correct final verdicts. Their position-selective self-distillation method withholds supervision at tokens where feedback sharply narrows the model toward one expression, improving generalization and outperforming outcome-supervised reinforcement learning by 2 to 9 percentage points on evaluated subjective categories.

Adaptive routing helps one language model handle mixed request types

Chen et al. introduce Adaptive-GEPA, a reflective optimizer that divides heterogeneous requests among learned specialist programs while improving those programs through execution feedback. In a Qwen3-8B experiment, it recovered the four underlying task families without being given family labels and raised the family-mean test score from 52.6 to 70.6.

Training on self-generated algorithms helps design agents escape performance plateaus

Lu et al. address a recurring problem in evolutionary algorithm-design agents: repeatedly refining solutions in context can stall when the task requires specialized knowledge. Their PCPO method instead trains the model on a curated population of its own solutions, outperforming frozen-model evolutionary approaches across chip-placement cases and producing substantial GPU-kernel speedups.

Adaptive agent teams improve translation across entire television series

Jin et al. developed SMART, a multi-agent subtitle translation system that adapts its workflow to each series without changing the underlying language models. The system achieved the lowest translation-error penalty across all 15 directions in a new long-form benchmark and led both automated and human evaluations on the public MuSC benchmark.

Language-Based Memory Helps Driving Models Track Earlier Road Events

Yan and colleagues introduce AD-Memo, a vision-language-action driving agent that converts selected observations into textual memory rather than repeatedly processing a long sequence of camera frames. Across all-way-stop and general-driving evaluations, the authors report better driving and scene question answering, while showing that the generated memories can be consumed by models that did not create them.

Reverse scoring helps diffusion agents escape repeated failed actions

Qiu et al. identify a decoding pathology in diffusion language model agents: after an action fails, adaptive masked decoding can favor repeating that same action instead of responding to the failure. They propose a training-free reverse score and report improved task success and progression across four multi-turn embodied benchmarks.

An LLM agent assembles sample-based quantum chemistry workflows

Tiwari and Ray present SQD-Agent, a framework for constructing quantum chemistry workflows based on Sample-Based Quantum Diagonalization, or SQD, from natural-language instructions. The system coordinates quantum circuits, classical solvers and analysis tools, and the authors demonstrate it on molecular energy-estimation tasks spanning four molecules, including an error-mitigation study on quantum hardware.

Cross-checking coding agents improves reliability in automated model research

Chen and colleagues present RankEvolve, a harness that automates repeated proposal, implementation, training and evaluation of ranking-model changes while enforcing a predefined experimental process. Their central result is that combining Claude Code and Codex as cross-checking execution nodes raised fully correct execution from 45.8 percent to 62.5 percent, although consequential defects still passed silently in 10.4 percent of cases.

Evolved prover interfaces make AI theorem proving cheaper and faster

Viennot, Baudart, and Lelarge treat the interface between an AI agent and a proof assistant as an object to optimize rather than fixed infrastructure. Their evolutionary process retained model-proposed tools only when they improved smaller agents, producing rocq-mcp-evolve, an MCP server that outperformed both a compiler-only interface and an established Rocq server on held-out theorem-proving tasks.

Cross-fitted feedback curbs shared errors in self-evolving search agents

Chen and colleagues identify “co-cheating” in self-evolving search agents: question proposers and solvers learn to agree on incorrect answers, making internal reward rise without comparable gains in factual correctness. Their CrossFit method breaks this shared feedback ancestry, roughly halving false agreement and raising average downstream scores by more than eight points.

Coding-agent approvals can authorize actions different from those executed

Wang tests whether an AI coding harness executes precisely the action covered by a human approval, rather than merely presenting an approval checkpoint. Instrumented experiments in Claude Code reproduce six classes of approval-to-execution mismatch, while a cryptographic token blocks identity-related failures but cannot detect changes that occur below the tool-dispatch boundary.

Clinical agents learn to gather evidence across longitudinal MRI records

Shao and colleagues built an interactive benchmark in which agents must investigate patient records and MRI scans from multiple visits before answering clinical questions. Their Qwen3-VL-8B agent, trained through demonstrations and reinforcement learning, achieved higher answer accuracy than substantially larger frontier models under the authors’ matched evaluation conditions.