Sean Spicer said Tuesday that “safe havens” in Syria are “at the top of the president’s foreign policy agenda.” Indeed, President Trump has advocated the establishment of “safe zones” within Syria since his campaign. There are many obvious problems with such a policy. It would probably require buy-in from Russia, Turkey and possibly the Syrian government, all of whom have raised objections. Beyond these diplomatic challenges, their establishment may violate the United Nations charter.

Establishing and protecting safe zones is likely to be enormously expensive, particularly if it involves ground forces. Despite Trump’s claims that the Saudis and other oil-rich Gulf states will foot the bill, they have made no commitments to do so. Given the fragmentation of territorial control in Syria, it is unclear where the Trump administration intends to establish these safe zones. Regardless of where they are established, the fragmented and fluid lines of control will likely make it difficult to prevent infiltration by armed groups, including the Islamic State.

Our recently published research raises another disturbing possibility: the establishment of safe zones for refugees in one part of Syria could make civilian victimization by rebel groups, such as the Islamic State, even worse in other areas of the country. Attempts at establishing humanitarian safe zones have failed, sometimes with terrible consequences, as in the 1995 massacre at Srebrenica. More civilians could be put at risk by safe zones purportedly established to protect them.

When rebel groups have access to some form of territorial sanctuary from rival forces (whether in the form of safe zones or foreign bases), they kill many more civilians on average than rebels without external sanctuaries. Both safe zones and foreign bases have similar effects on rebel groups: They provide a place free from government or rival forces where rebels can recuperate, recruit and train fighters, use as a staging ground for military operations, propagandize and expand their resource base. In short, safe zones increase the rebel groups’ lethality.

At the same time, safe zones or foreign bases often effectively sever the connection between the rebel groups and the civilians they claim to be fighting for. Rebels no longer have to compete with the government to win civilian “hearts and minds” to survive and fight. Rather than protecting civilians, they prey on them (even when those civilians make up their theoretical base of support).

Rebels with external sanctuary become, as economist Mancur Olson coined them, “roving bandits”: able to simply take what they need by force from civilians and return to their protected safe zone or sanctuary. For instance, when rebel groups such as the Sudanese People’s Liberation Movement (SPLM) and the Liberation Tigers of Tamil Eelam (LTTE) were primarily based outside the theater of conflict (in Ethiopia and India, respectively) there were periods characterized by high civilian victimization and predation. This was because these insurgencies lacked any incentive at the time to cultivate civilian support through governance or social service provision.

The results of our statistical analysis show that rebel groups with only external territorial control (e.g., safe zones, foreign bases, or refugee camps) — an increasingly realistic scenario in Syria as both the Islamic State and non-Islamic State rebels continue to lose territory in contested zones to government forces backed by Russian air power — kill, on average, twice as many civilians as those that control territory within the theater of conflict and lack access to sanctuary.

The Kurdish Workers’ Party (PKK) is another rebel group that benefited directly from a safe zone created in a foreign territory. The northern Iraq no-fly zone was established in March 1991 to protect Iraqi Kurds from Saddam Hussein and was unrelated to the Turkey-PKK conflict. The safe zone created a de facto independent Iraqi Kurdistan, providing sanctuary to the PKK.

With no government forces on the ground, the PKK quickly established bases in Iraqi Kurdistan that it used to launch a widespread campaign of attacks on civilian and government targets in Turkey. Our research leveraged this quasi-natural experiment to test the effect of the implementation of the no-fly zone on the PKK’s victimization of civilians.

We found that PKK civilian casualties — as well as the number of attacks on civilians — spiked dramatically after the 1991 establishment of the northern Iraq NFZ. Using data from the Global Terrorism Database, we found that in 1990 the PKK committed fewer than 50 attacks on unarmed targets. In 1992, the PKK committed more than 150 attacks on unarmed targets. This dramatic increase subsided only after a major Turkish counterinsurgency offensive culminating in an invasion of Iraqi Kurdistan in 1995.

Should a similar safe zone be established in Syria, particularly near Islamic State territory, it would be all too easy for a rebel group such as the Islamic State to take advantage of this new sanctuary. The Islamic State will not be flying planes over the territory to be shot down, nor does it rely on conventional ground forces. Even if ground troops were the solution, the Pentagon estimated in 2016 that securing safe zones would take between 15,000 to 30,000 troops, perhaps inviting a ground war between the United States and the Islamic State — precisely what the Islamic State wants.

Even tens of thousands of American boots on the ground might not guarantee that rebels would be blocked from entering a refugee safe zone. The French government’s administration of a Rwandan “safe zone” in 1994 should serve as a clear and terrifying reminder: members of the interahamwe, the perpetrators behind the Rwandan genocide, used these zones to continue the genocide.

What this ultimately means is that if the Trump administration is sincerely interested in establishing safe zones within Syria as a solution to the ongoing humanitarian catastrophe there, our research should give them pause. Although many others have pointed out that safe zones are impractical and unlikely to be effective, our research suggests that their establishment could, perversely, put even more Syrian civilians at greater and not lesser risk of victimization by the Islamic State and other armed groups.

Yu-Ming Liou (@YuMingLiou) is a PhD Candidate in international relations at Georgetown University.

 Megan A. Stewart (@Megan_A_Stewart) is an Assistant Professor of International Security at American University’s School of International Service and a Post-Doctoral Research Associate at the University of Virginia.