U.S., UK safety institutes release first joint assessment of generative AI model

The AI safety institutes of the United States and United Kingdom have released first-time joint findings from an assessment of a generative artificial intelligence model, showing that its ability to improve against deliberate misuse is slightly better than a “best performing” reference model.

The assessment signals future joint research by the U.S. and UK as part of an initiative announced earlier this year for international collaboration on developing methods for testing the safety and security of AI models before being...