The nightmare scenario for A.I. isn’t just killer robots … it could be nukes too.

A.I. ethicist Tristan Harris says the warning signs are already here … the Center for Humane Technology cofounder joined “TMZ Live” Wednesday, pointing to roughly 1,200 A.I. agents organizing themselves, creating a secret language, and even their own message board.

Harris is urging people to pressure lawmakers and top officials to pump the brakes on advanced A.I. … before the warning signs turn into something much worse.

As we reported … Anthropic researcher Jacob Coxon recently quit after warning humanity could be toast by the end of the decade if A.I. keeps advancing unchecked.

Harris says the U.S. and China need to stop treating A.I. like an arms race … because beating the other guy to Skynet still means everybody loses.

And this is where the nuke nightmare comes in … Harvey asked if A.I. could get into America’s nuclear systems … and Harris said that’s a real concern, pointing to war-game studies where A.I. repeatedly chose to launch nukes.

Harris says we’ve already seen evidence of A.I. going rogue … the question now is whether humans can stop it before things get worse.