THE CRUNCH

US intelligence agencies have issued a warning that China is systematically extracting capabilities from frontier AI models. The agencies describe the technique as a form of attack known as distillation, which involves capturing outputs and reasoning processes to train a different model.

US intelligence agencies have warned that China is systematically extracting capabilities from frontier AI models. The agencies describe the technique as a form of attack known as distillation, which involves capturing outputs and reasoning processes to train a different model.

WHAT HAPPENS NEXT

The agencies' warning suggests a need for enhanced safeguards against reverse-engineering techniques to protect the intellectual property and strategic advantages of frontier AI models.