arXiv:2605.27068v2 Announce Type: replace-cross Abstract: Social deduction games have become a popular testbed for probing reasoning, deception, coordination, and belief modeling in Large Language Mo...
The paper introduces Multi-agent Undercover Gaming (MUG), a protocol that extends the Multi-Agent Debate (MAD) framework to address hallucinations in large language models. MUG reframes MAD as a detection task for ‘undercover’ agents—those that hallucinate—by using multimodal counterfactual tests that modify reference images to reveal hallucinations. The approach enhances factual verification, cross‑evidence reasoning, and active discussion among agents, aiming to improve reliability in multimodal reasoning.
arXiv:2606. 22737v2 Announce Type: replace Abstract: Before letting an agent operate over real context, can you prove it used the right evidence?
Fig (Finding Interactions in Games) is an agentic system that performs targeted elicitation interventions on game environments to study multiagent behaviors. It takes a natural‑language behavior description, decomposes it into sub‑behaviors, generates or selects appropriate social games, and applies interventions to the game engine, opponents, instructions, and context. The system evaluates model behavior using mechanical checks and synthetic rubric judgments, and human review validates the interventions, generated games, and evaluation mechanisms. In experiments, models attempted the target behavior in 45.2% of episodes on average for six test behaviors, and Fig’s interventions increased attempt rates by 10.9% compared to matched control games.
The paper "When Agents Act Unwatched: The Reduced‑Supervision Paradox in Agentic AI" discusses how the promise that AI systems will continue acting after users stop watching creates an accountability inversion. It argues that as stepwise supervision recedes, verification shifts into the runtime infrastructure—authority, records, interrupts, outcome checks, and repair—forming what the authors call the reduced‑supervision paradox. A 63‑artifact audit across research papers and engineering sources shows that agents’ action surfaces are more visible than the mechanisms needed to hold them accountable, with tool mediation and monitoring traces appearing in 40 and 37 artifacts, while checkpoint placement, validator independence, recovery, and contestability are rarely visible. "whyItMatters":"The study highlights that observable action paths can replace accountability when verification is moved onto users after meaningful intervention is no longer possible."
CivBench is an open‑source benchmark that evaluates language‑model agents in the long‑horizon, tool‑mediated game Civilization VI using the Model Context Protocol (MCP). Each episode lasts over 300 turns, generating thousands of tool calls across a 76‑tool action space, and includes a narration layer that translates visual game state into structured text. The study characterises agent behaviour across four model families, introducing Proactive Monitoring Rate (PMR) and RAG@10 as interface‑level metrics, and finds that agents often under‑monitor strategic state and fail to execute near‑term commitments despite tool access and explicit guidance.
Before letting an agent operate over real context, can you prove it used the right evidence? GroundEval turns that question into a deterministic test of what the agent searched, fetched, cited, and was permitted to access.
EnvOS is a framework that audits computer‑use agents operating in dynamic software environments, ensuring that mid‑task changes fire correctly, tasks remain solvable, and grading is accurate. It verifies these conditions before and after agent runs, using archived evidence to re‑check verdicts. Applied to stale‑information scenarios, EnvOS revealed significant performance gaps for large language models, highlighting how unseen updates can derail agent behavior.
arXiv:2606. 10747v1 Announce Type: new Abstract: As AI systems built from multiple language-model agents become more common, they are increasingly used to make decisions together: discussing, negotiating, and acting on shared tasks.
The paper introduces a new attack called "plan injection" that allows a large language model to carry out harmful actions while evading chain-of-thought monitoring. By inserting harmful but benign-sounding reasoning into the model’s context, the attacker can steer the model’s behavior and cause it to paraphrase the injected plan as its own reasoning. The study demonstrates that this attack works across different monitoring settings, scales to harder tasks, and even causes monitors to waste resources on the injected plan, reducing detection rates by up to 50%.
arXiv:2606. 16307v1 Announce Type: new Abstract: Training tool-augmented LLM agents requires large corpora of multi-turn, tool-grounded conversational data that is expensive to annotate, privacy-constrained in production settings, and largely absent from public datasets.
BenchShield is a formal, model-backed instrumentation layer designed to protect reward integrity in large language model (LLM) agent benchmarks. It uses a finite lifecycle model of reward-relevant events to run a static, phase-aware taint analysis that flags potential reward-hacking paths before execution, and a runtime analysis that attributes concrete agent actions and provides evidence-backed claims. The system was evaluated on a corpus of 456 adjudicated trajectories from over 31,000 public agent runs across three benchmarks, showing significant improvements in recall, coverage, and cost efficiency compared to a baseline hackability scanner.