White-Box Agent Monitoring

Signal score68Credible early signal

Evidence23 / 50

Strategic45 / 50

StageEmerging

The signal has repeated beyond its initial observation: 1 observed days, 2 publications, 2 sources, and 2 qualified lifecycle layers.

Observation history1 observed days

First detected 5 days ago · seen 1 times this week.

First publishedNot yet published

This movement is still being watched for stronger evidence.

Observation history

How this signal developed

Each entry is a stored observation of the same market movement. Scores, stages, and evidence totals reflect what was known on that date.

October 9, 2026Analyst observation

Agent Monitors Look Inside the Model

First detected

Research on activation probes and a new commercial monitor both inspect internal model state instead of asking a second model to review every action. The mechanism could lower oversight cost and catch deception that never appears in a reasoning trace. It remains an early signal because the commercial performance claims are vendor-reported and the research has not yet established reliable production thresholds.

EmergingScore 682 publications2 sources

Signal lifecycle

How the market is forming

This lifecycle uses the 2 publications linked across the complete observation history.

2 of 3 market layers detected2 publications · 2 sources · 2 of 3 market layers

Detected

Creation

1 publication1 source

A new technology, term, or technical capability begins to appear.

arXiv

Detected

Product building

1 publication1 source

Builders and founders begin creating products around the idea.

bluesky global

No observations

Adoption

No evidence yet

Direct evidence shows usage, deployment, or real user friction.

Evidence

Why this signal appeared

These publications support the signal. The relevance score indicates how closely each item matches its subject.

arXivRelevance 90

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

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 datas...

Open source

bluesky globalRelevance 90

Goodfire says its new ‘inside-out’ monitors catch rogue AI agents at a fraction of the cost

Goodfire says its new ‘inside-out’ monitors catch rogue AI agents at a fraction of the cost Goodfire just launched what it says is a cheaper way to keep AI agents in check: instead of paying a second AI to read everything an agent does, its monitors peek insid...

Open source