Homeland Security is testing AI to help with immigration, trafficking investigations, and disaster relief

Photo by John Moore/Getty Images

The Department of Homeland Security (DHS) is rolling retired 3 $5 cardinal AI aviator programs crossed 3 of its agencies, The New York Times reports. Through concern with OpenAI, Anthropic, and Meta, DHS volition trial retired AI models to assistance its agents with a wide array of tasks, including investigating kid enactment maltreatment materials, grooming migration officials, and creating catastrophe alleviation plans.

As portion of the AI pilot, the Federal Emergency Management Agency (FEMA) volition usage generative AI to streamline the hazard mitigation readying process for section governments. Homeland Security Investigations (HSI) — the bureau wrong Immigration and Customs Enforcement (ICE) that investigates kid exploitation, quality trafficking, and cause smuggling — volition usage ample connection models to rapidly hunt done immense stores of information and summarize its investigative reports. And US Citizenship and Immigration Services (USCIS), the bureau that conducts introductory screenings for asylum seekers, volition usage chatbots to bid officers.

DHS’s announcement is scant connected details, but the Times study provides a fewer examples of what these pilots whitethorn look similar successful practice. According to the Times, USCIS asylum agents volition usage chatbots to behaviour mock interviews with asylum seekers. HSI investigators, meanwhile, volition beryllium capable to much rapidly hunt its interior databases for details connected suspects, which DHS claims could “lead to increases successful detection of fentanyl-related networks” and “aid successful recognition of perpetrators and victims of kid exploitation crimes.”

To execute this, DHS is gathering up an “AI corps” of astatine slightest 50 people. In February, DHS Secretary Alejandro Mayorkas traveled to Mountain View, California — famously the office of Google — to enlistee AI talent, and wooed imaginable candidates by stressing that the section is “incredibly” unfastened to distant workers.

Hiring capable AI experts isn’t DHS’s lone hurdle. As the Times notes, DHS’s usage of AI hasn’t ever been successful, and agents person antecedently been tricked into investigations by AI-generated deepfakes. A February report from the Government Accountability Office, which looked into 2 AI usage cases wrong the department, recovered that DHS hadn’t utilized reliable information for 1 investigation. Another lawsuit hadn’t relied connected AI astatine all, contempt DHS claiming it had. Outside of DHS, determination are plentifulness of documented cases of ChatGPT spitting retired mendacious results, including an lawsuit successful which a lawyer submitted a little citing nonexistent cases that the AI exemplary had completely made up.

Still, this enlargement isn’t DHS’s archetypal foray into AI. Some of the surveillance towers Customs and Border Protection (CBP) uses to show the US-Mexico border, specified arsenic those made by Anduril, usage AI systems to observe and way “objects of interest” arsenic they determination crossed the rugged terrain of the borderlands. CBP hopes to fully integrate its web of surveillance towers done AI by 2034. The bureau besides plans to usage AI to show authoritative borderline crossing zones. Last year, CBP awarded a $16 cardinal declaration to a tech and question institution founded by its erstwhile commissioner, Kevin McAleenan, to build an AI instrumentality that volition scan for fentanyl astatine ports of entry.

The caller DHS AI aviator programs, however, volition trust connected ample connection models alternatively than representation recognition, and volition mostly beryllium utilized successful the interior of the state alternatively than astatine the border. DHS volition study connected the results of the aviator by the extremity of the year.