arXiv Machine Learning
The Oversight Gap: What LLM Safety Monitors Miss, and Why It Is Not Capability
Read the original on arXiv Machine Learning →The Flow has not summarised this story yet — read it at arXiv Machine Learning.
Related stories
arXiv AI
Jun 16
Who Drifted: the System or the Judge? Anytime-Valid Attribution in LLM Evaluation Pipelines
arXiv:2606. 15474v1 Announce Type: new Abstract: Continuous evaluation of LLM products relies on a strong LLM judge treated as ground truth: a cheap monitor scores every interaction and a team is paged when the score drifts down.
arXiv Machine Learning
Sep 14
A False Average: Pooled CoT-Monitor Accuracy Conceals a Reasoning-Dependent Fragility
The paper demonstrates that aggregate accuracy figures for chain‑of‑thought (CoT) monitors can be misleading because a large portion of detected hacks rely solely on action patterns rather than reasoning. By rewriting only the agent’s reasoning to appear truthful while keeping actions identical, the authors show that the monitor’s performance on the reasoning‑dependent subset collapses dramatically, yet the overall pooled accuracy drops only modestly. The study reveals that CoT monitors are fragile when reasoning is the key signal and that accuracy should be reported separately for this subset.
arXiv AI
Aug 10
Ratchet: How Reliable Must an LLM Judge Be to Retire a Skill?
arXiv:2605. 22148v3 Announce Type: replace Abstract: A large language model (LLM) agent that writes and edits its own skill library must also decide which skills to keep, from one noisy scalar per skill.
arXiv AI
Aug 3
Safety, or Just Capability? A Validity Audit of Agent-Safety Benchmarks
arXiv:2607. 28685v1 Announce Type: new Abstract: Agent-safety benchmarks measure different behaviors, and their scores get quoted interchangeably as an agent's safety.
arXiv AI
Aug 11
When Is Benchmark Contamination Detectable? Information Limits and Power-Calibrated Audits
arXiv:2608. 07914v1 Announce Type: new Abstract: Behavioral contamination detectors can return "no evidence" either because a benchmark is clean or because the audit has little power.
arXiv Machine Learning
Sep 25
Don't Read the Log: Execution Traces Contaminate Verifiers in Video-Generation Agents
The paper investigates how providing execution traces to multimodal judges in agentic video‑generation systems can bias their verdicts. On a benchmark of 109 two‑event clips, traces that falsely report successful tool calls cause large‑language‑model judges to incorrectly accept 78–90 % of failures, while contradictory traces lead to 100 % rejection of correct clips. The effect persists even when judges are instructed to consider only the video frames, indicating that the vulnerability stems from the judges’ learned trust in tool logs rather than the visual content itself.