US Immigration and Customs Enforcement (ICE) faced a big problem: its artificial intelligence tool mistakenly sent many new recruits to field offices without the right training. ICE was in a hurry to add 10,000 new officers to its force by 2025. The AI tool was supposed to identify applicants with law enforcement experience to place them in a shorter, four-week online training program. Those without police backgrounds had to take an eight-week in-person course at the Federal Law Enforcement Training Center (FLETC) in Georgia. However, the AI scanner took any résumé with the word “officer” and put those people into the short training, even if they were not actual law enforcement officers. For example, ‘‘compliance officers’’ or applicants who only wanted to become ICE officers were wrongly included. This led to recruits working in the field without enough training. Officials said ICE field offices tried to give extra training before sending these officers out, but the error was serious. The problem was found in mid-fall, more than a month after the recruitment surge started. ICE quickly began fixing the mistake by reviewing résumés manually and sending wrongly categorized officers back to FLETC for proper training. It is not clear how many officers were affected or how many started immigration arrests without full training. This issue adds to concerns as ICE steps up its enforcement in US cities, including Minneapolis, where over 2,000 new officers have been deployed since late November. ICE offered $50,000 signing bonuses and aimed to meet Congress’s target of 10,000 new officers by 2025. However, due to the training errors, the officials said ICE did not fully reach the goal in actual officers working on the streets. The Department of Homeland Security has not commented on the issue.