The race to keep AI under control

Safety breaches are raising doubts about the ability of leading labs to keep increasingly powerful systems in check