President Trump is establishing a reputation for himself at the Threatener-in-Chief. Trump has said that if Congress won’t fund his border wall, he will shut down the federal government—a threat he quickly backpedaled. He has declared his own finances a “redline” in Robert Mueller’s Russia investigation, vaguely threatening to fire him, or Attorney General Jeff Sessions, if the investigation gets too personal. And then there’s North Korea. As Hurricane Harvey ravaged the Gulf Coast, 6,000 miles away, Kim Jong Un sent a ballistic missile flying right over Japan, America’s treaty ally, and followed that up with its sixth nuclear test. It was a slap in the face to both the president and his secretary of state, who had publicly praised North Korea for its “restraint” just days before. And it was a stark reminder that the North Korean nuclear problem is nowhere close to blowing over.

America’s North Korea policy is failing. It’s been failing for years, across several presidents. But the risk of conflict has grown dramatically in this administration—in part because Trump has gotten himself into a public threat war with the world’s most unpredictable and uncontrollable bully, and Trump’s go-to play is to threaten that man more.

Trump is committing deterrence malpractice—in four ways.

The first is making threats so obviously hollow that many of his own advisers don’t believe or support them. First there was the tweet about not letting North Korea get a nuclear weapon that could reach the United States. Kim has now tested a long-range missile capable of hitting Los Angeles and possibly reaching as far as Chicago, and claims the weapon it just tested could fit on such a missile. (U.S. intelligence officials have indeed assessed that Pyongyang has succeeded in the difficult task of miniaturizing a nuclear warhead.) Trump at that time warned that he was “sending an armada” steaming toward the Korean peninsula—except that the “armada” was actually steaming in the opposite direction for a pre-planned exercise with the Australian Navy. More recently Trump channeled his inner Kim Jong Un to warn the real Kim Jong Un that North Korea would be “met with fire and fury like the world has never seen” if it kept threatening the United States. So Kim made some actual fire and fury.

George Shultz likes to say he learned one of his most important diplomatic lessons when he was a young Marine Corps recruit: Never point your rifle unless you intend to pull the trigger. The one iron law of deterrence theory, the one point on which nearly all scholars and policymakers agree, is the importance of credibility. The more you have it, the more you can coerce successfully—getting others to do what you want them to do. Undermine your own credibility with one, and you risk undermining your credibility with many. In Game of Thrones, nobody believes Theon Greyjoy will do what he promises.  Everybody believes Ramsay Bolton will.  The same is true in real-life contests for power across borders. Researchers have found that countries will go to extraordinary lengths to preserve their credibility, even fighting wars they know they will lose, so that others in the future know they mean business.

Trump’s second form of deterrence malpractice is that he conflates power with influence. Sure, the U.S. is orders of magnitude more powerful than the Hermit Kingdom on every dimension. America ranks among the richest nations in the world. North Korea is among the poorest. America created the internet. North Korea has 28 websites. America has the world’s most powerful military and spends more on defense annually than the next several nations combined. North Korea has somewhere around two dozen nuclear weapons and a starving populace. America has now-troubled but historically strong alliances and relationships with more nations around the world than any other country. North Korea has half a friend: Beijing. But history tells us that sheer power is often not enough: The most powerful side in a contest of threats frequently doesn’t win. The U.S. had more military, economic, and political power than Japan in 1941, Vietnam in 1965, Iraq and North Korea in the 1990s, and Pakistan after 9/11. And yet, American leaders were unsuccessful in preventing Japan’s entry into World War II, gaining victory in Vietnam, keeping North Korea from leaving the nuclear nonproliferation treaty, convincing Iraq to withdraw from Kuwait without resorting to war, or winning Pakistan’s full assistance in countering terrorism, including the hunt for Osama bin Laden.

Nobel Laureate Tom Schelling famously called coercion, “the diplomacy of violence.” Having the capacity to inflict violence is just the half of it. The easier half. The diplomacy part is harder.

Which leads to Trump’s third type of deterrence malpractice: He talks too much. Effective deterrence is about signaling—often without words—that you really do mean what you say. In foreign policy, talk isn’t just cheap. It’s dangerous. In any conflict, all sides want to make it look like they’ll go to war even when they really don’t want to. The incentive to bluff is high and information (about the other side’s interests, resolve, and capabilities) is imperfect. Wars often occur through miscalculation and misperceived signals. Credible threats are hard to make with words alone.

All parents know this. When my three kids were little, my husband and I would take them to dinner in a restaurant about once a week to practice restaurant etiquette (old southern habits die hard). Each trip to “manners camp” would start off with a gentle reminder that good mannered kids got to stay for dessert while the others would go home early. Even at ages 3, 6, and 8, the kids saw right through that parental cheap talk: We all drove in one car, so there was no way we could selectively send one child home without punishing the others. Then one day, after teaching a class on Schelling, I got an idea. “Kids, tonight we’re taking two cars to the restaurant.” They got it immediately. I never had to say, “Yes, we really mean it. No manners, no dessert.” The second car did the talking for me. It was the most credible commitment signal I could send.

Throughout the Cold War, nuclear deterrence theorists reasoned that costly moves were the key to making threats more credible. The chief challenge was convincing the Soviets that America actually would annihilate the world with nuclear weapons if attacked. The threat of Armageddon, in turn, would keep the Soviets from striking in the first place. The only winnable nuclear war was a nuclear war that never happened.

American presidents, from Truman to Reagan, all used the same basic approach: using actions to speak more loudly than words. Ever wonder why there are so many U.S. forces in Germany— which numbered as many as 200,000 during the Cold War? These “tripwire” forces were stationed there to die, not just fight. American military planners were worried the Soviets might not believe the U.S. would extend its nuclear umbrella over NATO in an actual conflict, and as result, that the Soviets would be tempted to use their overwhelming conventional forces to invade. But if thousands of Americans died when Soviet tanks rolled across Germany, the U.S. would have no choice. American involvement, including the use of nuclear weapons, would be guaranteed. And mutual assured destruction would hang in the balance. This costly signal made some sense in an insane, Dr. Strangelove sort of way. American leaders didn’t need to say much. American troops said it all.

Making credible threats has gotten much harder as we have moved from the Cold War nuclear world of mutual assured destruction to a more complex threat landscape with cyber weapons that can inflict massive, asymmetric disruption. What makes threats credible today? I put this question to 250 foreign military officers over the past two years. Their answers were surprising. The No. 1 ranked factor in threat credibility was not a country’s willingness to risk soldiers’ lives, its superior power, or even its reputation for carrying out threats. It was domestic political support for military action.

If these survey results are any indication, President Trump is committing deterrence malpractice in a fourth way—by dividing the nation rather than uniting it, playing to our worst hatreds and his strongest base rather than bringing the country together in support of broader objectives that serve the national interest. Trump continues to hit all kinds of polling records—the bad kind. In April, marking his 100th day in office, he was the least popular president in modern times. Trump’s handling of recent events in Charlottesville has accelerated the erosion of his public support, even among Republicans. His current approval rating hovers at just 35 percent.

Domestic politics and foreign policy are intimately connected. President Trump entered office with a weak hand to deal with North Korea. But his approach to deterrence has only made it weaker.