← Back to feed

Caught in the Act: Probes Effectively Detect Sabotage and Catch Unverbalized Deception

L5 · ResearcherResearcharXiv· 10/8/2026

Advances AI safety monitoring techniques for detecting unverbalized deception in frontier models.

AI Summary

Researchers developed white-box probes that detect LLM deception with up to 99.7% AUC, outperforming text-monitoring baselines.

Excerpt

Recent incidents have highlighted the challenge of monitoring LLM agents and the danger of models deceiving people. We show that white-box deception detection via probes can be scaled up to frontier monitoring settings by collecting the largest deception dataset to date for training probes and introducing a novel probe architecture which can aggregate information across many layers and tokens. Our probes achieve 98.8% AUC in SHADE-Arena, surpassing an Opus 5.5 text-monitoring baseline, and show

Read Original
0 upvotes · 0 downvotes · 1 min read

Related Articles

L5 · ResearcherResearcharXiv
From Reactive Containment to Proactive Assurance: Lessons from OpenAI, Anthropic, and Google Agent Security Incidents

Analyzes 2026 agent security incidents from OpenAI, Anthropic, and Google to propose a proactive security framework with boundary assurance stack.

L5 · ResearcherResearcharXiv
A Balanced Data Diet: Addressing the Exploration Bottleneck in Mega-Scale RL for Robot Control

Researchers introduce Success Guided Sampling (SGS) to improve massive-scale reinforcement learning for robotics by focusing training on task configurations at the policy's capability frontier.

L5 · ResearcherResearcharXiv
CSF: Contextual Safety Filtering for Motion Generators

Researchers introduce CSF, a training-free safety filter that reduces danger events by up to 90% for text-conditioned motion generators while preserving benign motions.

L5 · ResearcherResearcharXiv
On the estimation and validity of AI time horizons---a statistical look at the METR plot

Researchers propose improved statistical methods for estimating AI time horizons using splines and item-response theory, revealing nonlinear relationships between human task completion time and AI difficulty.

L5 · ResearcherResearcharXiv
BrickBench: Evaluating Agentic Brick Design

Researchers introduce BrickBench, a benchmark for evaluating AI agents in designing physically buildable LEGO sets from text prompts.