Report / Tech
Published by NOISE
Frontier AI labs disclose few public plans for containing rogue models
From EVENING EDITION. The edition archive has expired; this report remains available.
A Guidelight AI Standards assessment found limited public evidence that leading labs have formal plans to contain models that subvert human control, with OpenAI scoring highest and Anthropic and Meta lowest. TechCrunch reported that the assessment covered publicly available information and may not reflect undisclosed internal safeguards. The findings come as California and New York requirements push developers to disclose responses to critical AI safety incidents.
Saved versions
These dates record changes to NOISE’s copy. They are not event dates.