This is not a trivial point. Selective pressures will push AI towards unsafe directions due to arms race dynamics between companies and between nations. The only way, other than global regulation, would be to be so far ahead that you can afford to be safe without threatening your own existence.