Dangerous incidents related to artificial intelligence reached record levels in 2024, but regulation is still lagging, writes Anthony Cuthbertson. Experts warn it could take a catastrophe before we take the risks seriously – but by then it may be too late