CS Paper Radar

DelegationBench: Measuring When AI Agents Should Ask Before Acting

Shiva Pochampally

cs.AI cs.CLcs.HCcs.LG ·submitted 2026-10-04 20:54 UTC · v1
benchmarkcode/data

Abstract

AI agents that send emails, edit files, and make purchases must decide when to act on their own and when to check with the user first. This decision is usually evaluated by showing a model a proposed action, asking whether it should proceed, and scoring agreement with human labels. We introduce DelegationBench to test whether such scores can be trusted. It has 156 scenarios with four possible responses (act, ask for permission, ask for missing information, refuse), and most scenarios come in matched pairs that change a single feature: whether the action was requested, what is at stake, whether it can be undone, or who will see it. Across ten models from five families, agreement scores mislead in three ways. A simple keyword rule, which we wrote after seeing the benchmark, agrees with our annotators more often than eight of the models, yet its decision changes in only 9 of 48 matched pairs. Equivalent ways of asking the same question change how often a model acts by up to 52.5 percentage points. And every model stops to ask the user less often when it must carry out the task with tools than when it judges a proposed action. When rules are stated explicitly, the same models follow them almost perfectly, so the gaps are not explained by a general inability to follow rules. We release the benchmark and evaluation tools and recommend reporting these properties separately rather than as one score.

Comment
35 pages, 7 figures. Code and data: https://github.com/PieLord757/delegation-bench

Closest papers

Nearest neighbours by meaning across the whole index.

Value-Sensitive Delegation in Everyday AI Agent Use: Evidence from OpenClaw

cs.HC cs.AI ·2026-09-18 ·Renkai Ma, Ruyuan Wan, Xuan Lu, Fan Yang, Chen Chen +1 more
similarity 0.69
Abstract

Users increasingly delegate work to autonomous AI agents, yet evaluations typically measure task completion rather than the values users prioritize. Using Value Sensitive Design, we analyzed, with LLM assistance, 73,093 first-person Reddit posts about using OpenClaw, each for its human value, agent aspect, value fulfillment, and user outcome. The 21 values form six value groups, including Autonomous, Dependable, and Affordable Operation, Bounded Reach, Reviewability, and Equitable Access. Relative to each aspect's corpus share, values clustered not at the agent's outputs but at the operating conditions users set around a run. Values were usually met where users described what the agent delivered, in five of six groups, and mostly unmet where users described supervising it, in all six groups. We conceptualize this pattern as value-sensitive delegation. Supporting human values requires attention not only to what an agent accomplishes, but to the conditions users set around delegation, including cost, access, and oversight.

0.17

Dependable Agents Need Hard and Fuzzy Rules

cs.SE cs.AI ·2026-10-31 ·Frederick Roth
position similarity 0.69
Abstract

Language models let software read requests in ordinary language, but they keep rules as people do: usually, and less so under pressure. Conventional code keeps rules exactly, but it cannot read a request. We argue that dependable agents need both, because the two kinds of rule cover overlapping and unequal regions of what an agent might do. Code holds the limits that can be stated exactly. Learned fuzzy rules cover the judgment that cannot. We test the hard layer in three domains, purchasing, delegation of authority and data access, on 1,340 episodes written by authors outside our team. Models left to decide for themselves executed unauthorized actions in up to 151 of 340 purchasing episodes and 269 of 470 delegation episodes, and the strongest model breached in every condition. A hard guard between proposal and action executed none, and an exhaustive search of 183,204 proposals agreed. In a further test, where an outsider wrote the owner's rules and a program mapped them to the guard, two episodes breached at a boundary we had written down in advance. We then test the pair. Concept Programming instills a named rule in a model and measures its errors on wording it never saw. A small model taught four judgment rules this way flagged 95 of 96 rule-breaking purchases that no exact rule can see, and 4 of 108 legitimate requests. Placed in front of the guard, it left 1 of 192 rule-breaking requests unstopped, where the guard alone left 77 and 70, and the guard still held every exact limit. A prompted frontier model did about as well, so the claim is that a small taught rule can do the job, not that it does it better.

0.25

Do User-Authored Permission Policies Improve Protection Against AI Agent Overreach?

cs.HC cs.CR ·2026-08-27 ·Ting Yan
publishedhighly cited similarity 0.68
Abstract · 15 pages, 5 figures, conditionally accepted at the 2027 ACM Conference on Intelligent User Interfaces (IUI 2027)

AI agents are poised to become a primary interface to digital products, acting across email, files, payments, and personal data. People without professional software backgrounds need understandable, reusable ways to control actions across services. We examine a mechanism in which a language model maps actions to plain-language consequence categories with user-authored "allow", "ask", or "never" rules. We ask what is gained and lost when decisions are made in advance as reusable rules rather than separately for each action. We analyzed 113 participants without professional software backgrounds across three conditions: per-action human-in-the-loop approval (HITL), automated per-action model review (AUTO), or user-authored consequence policy (POLICY). Participants judged 2 examples in each of 4 consequence categories; POLICY participants then set one rule per category. All supervised an 18-action simulated day, including 7 overreach actions. POLICY blocked less overreach than HITL (-20.1 percentage points, 95% CI [-32.1, -8.1]) and AUTO (-14.5 points, 95% CI [-25.8, -3.2]). POLICY lowered runtime prompts from 18.0 to 10.9, but total intervention time was not reliably lower when rule setup was included. Exploratory analysis showed that participants chose "ask" for 114 of 140 POLICY rules, returning most overreach actions to runtime. Of the 148 overreach actions executed in POLICY, 133 followed human approval and 15 ran automatically under "allow" rules. Across all 7 overreach actions, POLICY had the highest approval rate. Counterintuitively, user-authored rules did not by themselves provide stronger protection: many actions outside users' original requests went through after users approved them. These results reveal a gap between preference and commitment: repeatedly choosing "ask" preserves case-by-case choice but prevents a standing policy from settling decisions in advance.

0.65

JevAgentHarness: Benchmarking, Deploying and Training Decision Models That Share Control with LLMs

cs.AI ·2026-10-15 ·Kunlun Zhu, Tao Feng, Xu Han, Heng Wang, Ling Yang +3 more
code/datahighly citedcommunity pick similarity 0.67
Abstract · Code: https://github.com/JevAgentHarness/JevAgentHarness

Large language model (LLM) agents now fix software, operate computers and serve customers, but their harnesses, the programs that run them, call a large LLM at every step, which makes long tasks slow and costly. Most of these calls end in a multiple-choice decision, such as which tool to call next. Decision models such as Jev answer these questions directly, at a fraction of an LLM call's time and price. We ask how a harness can delegate these decisions to a decision model with little loss of success. To evaluate decision models in agentic tasks, we build AgentDecisionBench, which tests them from isolated questions up to control of a running agent, on public benchmarks and our DecisionGym environment. The decision model most accurate on isolated questions solved only 69.5% of agent tasks, and without a fitted confidence threshold, delegating tool choices lowered success by 19.0 points. Building on these findings, we then design JevAgentHarness, which fits this threshold for each choice on the agent's own trajectories and lets the LLM decide below it. Because existing decision models never see the states an agent reaches after its own mistakes, we train AgentDecision-4B, an open 4B model, on such states with Jev's answers as labels. On tasks unlike its training data, the agent solved 91.5% of them with AgentDecision-4B and 82.0% with Jev. On tasks like their training data, 4B models trained this way made agents 81.1-89.0% cheaper and up to 2.5 times faster, without losing 3 or more points of task success. The models, data, and harness will be released at https://github.com/JevAgentHarness/JevAgentHarness.

0.67

Grounded Checklist Partial Credit for Agent Skill Trajectories

cs.SE ·2026-08-26 ·Suliu Qin, Lu Yin, Xilu Wang
similarity 0.66
Abstract

Language-model agents increasingly tackle long-horizon tasks in interactive environments, yet their evaluation commonly relies on task-level success rates by reducing an entire execution trajectory to whether the task passes an official verifier. This binary score hides partial progress and is particularly limited for procedural agent skill evaluations, since a skill can alter execution without changing the final outcome. While checklists provide finer-grained evaluation by scoring individual task requirements, costly manual authoring and unreliable automatic generation make trustworthy evaluation difficult to scale. To address these challenges, we introduce Grounded Checklist Partial Credit (GCPC), a human-governed and LLM-instantiated partial-credit evaluation of agent trajectories. Humans define reusable rules once, from which an LLM instantiates a task-specific checklist grounded in the task instruction and official verifier. To keep judgment tied to evidence, a judge scores each item from execution log evidence alone and abstains when evidence is missing. A separate scripted step then applies the official verifier outcome to the score. Across a 4,455-trajectory, deduplicated SkillsBench evaluation population, GCPC better discriminates official PASS and FAIL outcomes than holistic judging on the shared subset (AUC 0.689 vs. 0.619). Human evaluation on 96 trajectories from 12 tasks shows that GCPC aligns more closely with human assessments of progress. Applied to 1,946 matched with/without-skill pairs, GCPC exposes the effects hidden by pass@1: among 879 pairs whose binary outcome does not change, 20.9% improve by more than 0.10 while 18.7% regress by the same margin. The GCPC pipeline also transfers to Terminal-Bench and SWE-bench, demonstrating applicability beyond skill-conditioned evaluation.

0.40

BekchiAI: Measuring, Observing, and Controlling LLM Agents in One Click

cs.AI ·2026-08-27 ·Mesut Toruk
benchmarkcode/data similarity 0.66
Abstract · 8 pages, 1 Figure, 6 tables

Large language model agents reason, call tools, and act autonomously over many steps, but their agentic skills-correctly sequencing tools, planning under dependencies, judging untrusted inputs, and grounding generated arguments-are hard to measure with accuracy-only leaderboards. We present BekchiAI, which addresses both sides: a benchmark for measuring agentic skill and a platform for observing and controlling live agents. The BekchiAI-Benchmark, a suite of 13 tool-using ReAct agents across 7 task categories (arithmetic, structured/SQL, security detection, URL grounding, planning, orchestration, and tool-policy), totalling 2,057 deterministic, committed test tasks. Every task is verifier-checkable gold answers are computed by running canonical SQL against a real database, computing the exact schedule of a directed acyclic graph (DAG), or evaluating closed-form lambdas including adversarial security samples paired with deliberately imperfect signature scanners so a score reflects the model's own judgment, not the copying of an oracle. We define a small set of behavioral metrics beyond accuracy-tool-call adherence, URL hallucination and source-match, and per-model token cost and report a four-model comparison (Qwen3.7-Max, gemma-4-31B-it, gemma4:26b, gpt-oss-120b) whose story is in the per-family spread, not the aggregate. The benchmark runs are executed using the provided evaluation scripts. BekchiAI-Platform is a complementary web-based observability and control layer for deployed agents, providing full token and latency telemetry as well as remote run termination. The benchmark, evaluation tools, and platform are publicly released.

0.32

Evidence-First Agentic Verification

cs.AI cs.LG ·2026-10-13 ·Reza Sanayei, Razvan-Gabriel Dumitru, Sunjie Hou, MohammadHossein Rezaei, Haris Riaz +5 more
benchmark similarity 0.66
Abstract

Long-horizon, semi-verifiable agentic tasks often rely on rubric-based evaluation, but the evidence needed to assess these tasks can be spread across the environment state and the execution trace. A judge may overlook this evidence or accept unsupported claims, including an agent's own reports of success. When rubric verdicts serve as reinforcement-learning rewards, such errors can reinforce undesirable behavior. We introduce Sauron, an evidence-first agentic verifier that investigates both state and trace and relies on a claim only after independently validating it against its source. Bounded queries and joint investigation of related criteria reduce context use and redundant work, and each verdict carries an auditable receipt of its supporting evidence. Across 2,586 human-labeled criteria from 100 personal-assistant episodes, Sauron achieves the highest F1 of five verifiers on all eight backbones tested while costing less than Gandalf, the strongest baseline. With GPT-5.6 Sol, it reaches 0.96 F1 versus Gandalf's 0.92 at 23% lower cost, with a wider margin on process criteria. On a paired fault-injection suite covering 13 families, several of which simulate reward hacking, Sauron detects 97% of injected failures while flagging only 2% of clean controls, outperforming baselines even when they are given the full trace. On TRAIL, a benchmark for detecting and localizing real agent errors, Sauron leads both Gandalf and TRAIL's own judge on all three quality metrics.

0.30

PACT: Can Enterprise AI Assistants Be Trusted Under Pressure?

cs.CL cs.AIcs.CYcs.LG ·2026-09-16 ·Mika Okamoto, Ansel Kaplan Erol
benchmarkcode/datahighly cited similarity 0.66
Abstract · 26 pages, 12 figures, 17 tables. Includes technical appendix; Dataset: https://huggingface.co/datasets/trace-ai-labs/pact; Code: https://git…

As corporate AI adoption continues to grow, enterprise-grade LLM agents are being deployed into sensitive contexts such as hiring, healthcare, and finance. In these contexts, compliance with rules specified in an agent's system context is a first-order legal concern. Currently, no evaluation framework systematically measures which LLM models tend to violate compliance rules, especially under pressure from a persistent user, a hurried manager, or circumstances where violation is convenient or attractive. We introduce PACT (Pressure-Applied Compliance Testing), a benchmark for rule-following under pressure in AI agents assisting employees in daily tasks across twelve regulated enterprise domains and forty-eight scenarios, each set in a realistic multi-turn conversation. Each benchmark item pairs a standing rule against a rule-violating shortcut, and applies a battery of pressures across different wordings and system-prompt modes. We construct PACT component by component under strict LLM-as-judge auditing to ensure samples are unambiguous, ungameable, and realistic enough to avoid eliciting evaluation-aware behavior. We use PACT to profile LLM compliance across six complementary metrics that create a holistic picture of an AI assistant's robustness under pressure and throughout multi-turn conversations, its transparency, and ability to correctly discern where a rule applies. We aggregate this profile into PACTScore, a reliability-weighted compliance rate over all items and modes. Our results across 22 common LLM models spanning multiple providers and sizes show substantial variability in compliance across models and metric dimensions. Even the strongest assistants mis-apply a rule on 6 to 10% of items, and ordinary user pressure raises the violation rate by 65% on average. PACT highlights compliance risks in LLM assistants, motivating guardrails and careful model selection.

0.63