The institute will analyse AI models' performance and the risks they might pose and ramp up information sharing with institutes in other countries.