US spending watchdogs have called on the National Weather Service (NWS) to deliver an updated plan for its AI language translation project to reduce the risk posed by extreme weather events to people not proficient in English.
The call from the Government Accountability Office (GAO) comes as vast swathes of the US are pinned down by a winter storm that has already claimed lives.
But the NWS AI translation effort has been caught between a drive to reach as many people as possible, administration changes, and a White House Executive order to cement the role of English as the country’s official language.
The GAO report [PDF] noted that 26 million people in the US have “limited ability to read, speak, write, or understand English.” Two-thirds of people with limited English speak Spanish, while other major languages include Chinese, Vietnamese, Korean, and Tagalog. French rounds out the top ten.
If they can’t understand the NWS’s emergency weather messages sent to mobile phones or via radio and TV, this can lead to deaths and injuries, economic effects, and “confusion,” for example, during evacuation efforts.
Text alerts, or WEAs, are already distributed in Spanish, while the Emergency Alert System (EAS) over radio and TV is English-only. Most other non-alert messaging is English-only too.
There are FCC requirements to support additional languages for WEAs by June 2028. At the same time, Executive Order 14224, signed on March 1, 2025, designated English as the official language of the US. This has caused uncertainty over exactly how far the NWS should go in extending its translated alerts.
It has also created uncertainty around the NWS experiments in using AI to provide translations of alerts.
This started in 2021, with “translated products on a public website since October 2023,” the report said. It added that as of December 2025, AI was used to “translate selected products issued by about a quarter of its weather forecast offices and the National Hurricane Center into five languages.”
The GAO said the NWS had contracted with a commercial vendor to develop its translation systems, “and relies on internal bilingual NWS staff to help train the AI models.” From 2022 to last year, it spent about $2.7 million on the project, which covered five languages.
Training models for a new language took about three months, with NWS bilingual staff and contractors reviewing translations. Once up and running, the “translation process generally takes about one minute and is integrated into the system that forecasters use to generate weather products.”
But the program was briefly paused as of March 2025 due to the need for department-level approval of National Oceanic and Atmospheric Administration (NOAA) contracts for more than $100,000.
A new deal was subsequently signed, but at a lower level, meaning plans to support further languages are on ice until at least April 2026. The report says that officials did not seek further funds for the AI project in 2026 as they were awaiting guidance on the implementation of the executive order.
The GAO said the AI program could extend the accessibility of “weather products” and reduce risks to public health and safety during extreme weather events.
But the watchdog said the NWS “could better manage the project by establishing and communicating measurable performance goals, resource needs, and strategies to address internal and external challenges.”
It called for an updated implementation plan to help “NWS use its existing resources to support the highest priority activities, realistically plan for the future of the project, and communicate resource needs to Congress.”
But the agency also highlighted the need for human input too as literal translations do not always capture nuance. For example, a brochure on rip tide safety “used corrientes de resaca for rip current, which for certain Spanish speakers translates to ‘hangover current.'” ®
