Vaibhav Sisinty, founder of GrowthSchool, warned that AI is smart enough to realise when it’s being tested for safety. He added that AI changes its behaviour during safety checks to appear more moral than it actually is. This means we might be building something dangerous without even knowing it, he added.