Push AI forward too fast, and catastrophe could occur. Too slow, and someone else less cautious could do it. Is there a safe course?

https://www.cold-takes.com/racing-through-a-minefield-the-ai-deployment-problem/