Correlated Failures of AI Approval Monitors
Before an AI agent takes a consequential action, such as paying, messaging or changing code, several monitors check it.
ProofPaper ↗
Key points
- Their verdicts are combined by aggregating scores or by counting approvals, whatever inputs the monitors share.
- Monitors that share a base model or a poisoned input, however, may fail together.
- We borrow the fail-prone systems of distributed computing: lists of monitor sets that may fail together.
- We measured twelve small open-weights monitors on payment approval under two pre-registered protocols.
Sources (1)
- [1]Correlated Failures of AI Approval MonitorsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 21, 04:09 AM
Before an AI agent takes a consequential action, such as paying, messaging or changing code, several monitors check it.
Their verdicts are combined by aggregating scores or by counting approvals, whatever inputs the monitors share.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 20, 2026Full Guide to OpenAI dots
- Oct 20, 2026[AINews] DeepSeek DeepInfer: Open Source SOTA Inference running 155T toks/day/cluster making $22m/day at ~80% margin for a 81 day payback
- Oct 20, 2026Frontier LLMs Enter the World of Robotics: A Cross-Model Evaluation on RoboColiseum
- Oct 20, 2026Exa 3.0 Search API Tops Perplexity, Brave and Parallel on 8 Benchmarks
- Oct 19, 2026MiniMax M3.1 Flash shows China's AI price race has no floor in sight
- Oct 19, 2026AI researchers stranded Claude on a virtual desert island. It started to “play.”