Online Safety Monitoring for LLMs

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Mona Schirmer, Metod Jazbec, Alexander Timans, Christian Naesseth, Maja Waldron, Eric Nalisnick
• Venue: arXiv: cs.AI
• Published: 2026-07-02
• Collected: 2026-07-03
• URL: http://arxiv.org/abs/2607.02510v1
• PDF: https://arxiv.org/pdf/2607.02510v1
• DOI: UNKNOWN

Plain-Language Brief

Researchers propose a simple online safety monitoring system for Large Language Models (LLMs). The system uses a verifier signal from an external model to decide when safety can no longer be assumed, by setting a threshold through risk control. In experiments on mathematical reasoning and red teaming datasets, this basic design performs competitively with more complex monitoring systems based on sequential hypothesis testing.

What the Document Says

• The research is about an online safety monitoring system for LLMs.
• The system uses a verifier signal from an external model and a threshold set by risk control to decide when safety is compromised.
• In experiments on mathematical reasoning and red teaming datasets, the basic design competes with advanced monitors based on sequential hypothesis testing.

Key Concepts Mentioned

• Large Language Models (LLMs)
• Online Safety Monitoring
• Verifier Signal
• Risk Control
• Thresholds
• Mathematical Reasoning
• Red Teaming Datasets
• Sequential Hypothesis Testing

Claims Made by the Document

• LLMs remain prone to unsafe outputs at deployment time.
• Online monitoring of LLM outputs is crucial when safety cannot be assumed.

Evidence Used

• Despite alignment training, LLMs remain prone to generating unsafe outputs at deployment time. Monitoring outputs online and raising an alarm when safety can no longer be assumed is therefore critical.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Despite alignment training, LLMs remain prone to generating unsafe outputs at deployment time. Monitoring outputs online and raising an alarm when safety can no longer be assumed is therefore critical. We study a simple real-time monitor that turns a verifier signal from an external model into an alarm decision by thresholding, with the threshold calibrated via risk control. In experiments on mathematical reasoning and red teaming datasets, we show that this simple design is competitive with more advanced monitors based on sequential hypothesis testing.