Guidelight finds frontier-lab control practices only partially implemented
Guidelight compared six publicly described frontier-lab control practices and found uneven implementation. The assessment is based on public evidence, not a full audit of internal systems.
A dated report connected to the underlying research where available.
The report
Guidelight assessed frontier AI labs against six control practices using public disclosures and externally visible evidence. The result was uneven implementation rather than a simple safe/unsafe split.
The methodological limit matters: public documentation can reveal whether a lab describes a control, publishes evidence for it or exposes a governance mechanism, but it cannot substitute for direct access to internal systems.
DiggingBeagle should therefore use the assessment as comparative evidence about disclosed control maturity, not as a complete security audit of hidden infrastructure.
DiggingBeagle. “Guidelight finds frontier-lab control practices only partially implemented.” https://diggingbeagle.com/news/guidelight-frontier-control-assessment-august-2026/
DiggingBeagle is a non profit research project documenting AI security incidents, agent failures, vulnerabilities and AI-assisted operations. A case keeps its claims beside the sources that support, contest or limit them. Later updates stay visible, so a reader can see when the account changed.
We publish case reconstructions, dated reporting and analysis across records. Each has a different evidentiary role. About the project and our methodology explain how the work is reviewed.