Homeland Security Secretary Kristi Noem, on Face the Nation on January 18, described people in ICE custody this way: "70% of them have committed or have charges against them on violent crimes" [1].
The claim contains a defensible kernel worth stating plainly. "Have committed or have charges against them" is a deliberately broad construction: it folds pending charges in with convictions, and, read one way, it invites the listener to hear every kind of criminal-legal contact at once. Read at its most generous, it is a statement about accusations as much as proven conduct.
The record does not support it even read generously. A Cato Institute analysis of ICE data from October and November 2025 found that 5 percent of detainees had a violent conviction, that 73 percent had no conviction of any kind, and that 47 percent had neither a conviction nor a pending charge [2].
Other trackers, as compiled by Poynter, land in the same territory, well short of 70 percent, and none of them measures violence. The New York Times put violent convictions at about 7 percent. TRAC found that 41 to 57 percent had any conviction at all. The Deportation Data Project found that about 64 percent had any charge or conviction [1]. The highest figure in that set, 64 percent, counts any charge or conviction of any severity, which is the opposite of the word Noem chose.
Here is the steel-man on the definitional point. Read "committed or have charges" as "any contact with the criminal-legal system, charged or convicted, of any severity," and some trackers do approach a majority. That reading is available in the syntax. What it cannot survive is the word "violent." On violence specifically, the datasets converge near 5 to 7 percent, an order of magnitude below the number Noem gave.
The distance between the two matters because the claim is doing work. It is the public case for the pace and reach of detention. When 73 percent of the people held have no conviction of any kind, describing the population as 70 percent violent is not a rounding difference. It is a portrait of a different group of people than the one the data describes.