Tech giants pledge AI safety commitments — including a ‘kill switch’ if they can’t mitigate risks