Skip to content

Topic

AI model monitoring

Systems that watch AI model behavior during training, evaluation or use in order to detect and respond to harmful or unauthorized actions.

Current clusters

security4 publishers

OpenAI Holds Its Largest Frontier Training Run Over Its Own Critical Cyber Threshold

A model OpenAI calls Astra may meet the Critical cybersecurity bar in its Preparedness Framework. The framework now appears to be costing the roadmap compute, schedule and money.

Perspective Coverage

4 publishers
Builder
Builder 34%
Operator
Operator 50%
Investor
Investor 16%

Reality

Evidence62
Adoption30
Hype gap+10
Incentives
Insufficient
Confidence58