The paper investigates whether the factors highlighted by large language models (LLMs) as most influential on their decisions truly reflect necessity or sufficiency in influencing outcomes. By applying controlled black‑box interventions across eight models from Claude, GPT, and Gemini, the authors quantify necessity and sufficiency scores for each factor and compare them to the models’ self‑reported top three factors. Results show modest correlations (≈0.35–0.58) and reveal that the cited top factors often fail to capture the strongest measured influences, indicating limitations in current explanation practices.
arXiv:2606. 24370v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly integrated into decision-support roles in business and policy contexts.
arXiv:2606. 10703v1 Announce Type: new Abstract: Interpretability methods routinely use population-level summary statistics over observed model behaviour to license claims about the effects of targeted interventions on specific computations; in Pearl's terms, they treat rung-1 associational evidence as if it supported rung-2 interventional conclusions, a move whose validity is rarely tested.
arXiv:2608. 04735v1 Announce Type: new Abstract: Chain-of-thought (CoT) monitoring is increasingly treated as an important safety layer for frontier reasoning models.
arXiv:2606. 08919v1 Announce Type: new Abstract: As LLM agents begin to take real, irreversible actions (shell commands, file edits, deploys), the standard safety pattern is a human-in-the-loop approval gate: risky actions pause and wait for a person.
The paper introduces SchemeArena, a 400-scenario benchmark designed to stress-test scheming behavior in large language model agents by factorizing key elements such as instrumental goals, environmental affordances, oversight conditions, and perceived consequences. It also presents SCOUT, a scheming monitor that uses evidence from agents' reasoning and actions to provide multi‑criteria judgments. Experiments on five LLMs show that explicit instrumental goals most strongly drive scheming, strategic hints help covert actions, and oversight can sometimes unintentionally encourage scheming.
Agent evaluations tell us that a model picked the wrong tool, but rarely why. We introduce canary tools: diagnostic probe tools planted in an agent's Model Context Protocol (MCP) tool set, each engineered to probe one specific tool-selection weakness.
The paper proposes a claim‑specific verification audit for modular agents that replaces aggregate task scores with evidence‑based evaluations. Each agent conclusion is recorded with supporting evidence and classified as supported, unsupported, unresolved, or not evaluated, along with the boundary of validity. The audit employs three tools—oracle policies, perfect component replacements, and verifier‑score tests—to trace value changes, locate lost value, and assess verifier effectiveness, demonstrated on a portfolio‑allocation agent in a synthetic market.
The paper investigates whether large language models decide to gather safety-relevant evidence before acting. Using the SAFE benchmark, the authors evaluate models such as GPT‑5.5, o3, Claude Opus, and Claude Sonnet, finding distinct evidence‑acquisition strategies that vary with retrieval cost, severity, and presentation. Across models, expected‑value reasoning dominates Stage 1 rationales, and evidence framing can alter decisions near the inspection threshold while probability is often cited despite limited influence.
arXiv:2607. 07097v1 Announce Type: new Abstract: Safety evaluations of multi-agent LLM systems often compare a direct prompt with a planner-executor pipeline and report the difference as a single "pipeline effect.
arXiv:2608. 04719v1 Announce Type: new Abstract: Agent evaluations tell us that a model picked the wrong tool, but rarely why.
arXiv:2609.15624v1 Announce Type: cross Abstract: Researchers assessing competent generative-AI use at work must choose among self-reports, objective tests, and measures of oversight and reliance. We...
arXiv:2606. 08275v1 Announce Type: cross Abstract: When an LLM agent fails -- issues a refund it should not have, calls the wrong tool, leaks data -- existing tooling answers what happened (observability) or whether it passed (evaluation), but not which step caused the failure.