Trump’s EPA Develops New AI Tool to Aid Staff, Not Decide Policy

The EPA is rolling out an artificial intelligence tool agency-wide on Thursday that it says will supercharge its staff’s capabilities, a goal the new administrator set when first joining the agency.

“We see it as a workforce multiplier,” said Dwane Young, the Environmental Protection Agency’s chief AI officer. “It will really help staff do a lot of the work that they are routinely doing much more quickly.”

All EPA employees will be encouraged to widely use the tool, which is based on the OpenAI platform but with an internally created “wrapper” around it, Young said.

EPA Administrator Lee Zeldin has promised to deploy AI to improve the agency’s technical capabilities for tasks such as reviewing public comments, overseeing grants, and reducing review backlogs.

The rollout of the new tool comes as the EPA has sought to reorganize and cut staff, part of a broader government restructuring under President Donald Trump.

But Carter Farmer, the EPA’s chief information officer, said the agency isn’t adopting AI to replace staff.

“There’s no shortage of work and there never will be,” Farmer said. “We have more work than we can shake a stick at. At the end of the day, our goal is to deliver on our mission and our promise to the taxpayers.”

Farmer also said AI will not be used to make policy decisions.

While the tool can be used to make a first pass at drafting an email or summarizing a document, everything it does must be reviewed by a human, Farmer said.

Daily Tasks

One example of what the tool can do is streamlining the review of total maximum daily load reports for water pollution, which can be up to 80 pages long.

When the EPA receives one of those reports, it then faces a short timeline to review each report, turn the information into data, and upload the data so the public can access it—all of which takes huge amounts of time, Young said.

He recently used the AI tool to summarize the state water reports, identifying the affected waters and pollutants, then creating an uploadable file.

It took “five minutes of prompt engineering, telling the tool what it was I expected it to do, copying and pasting the document into the tool, and then interrogating, asking it questions,” Young said, and the tool “did it almost perfectly.”

“There’s a few things that would need to be tweaked, but it’s an incredible time saving opportunity for an EPA staff person,” he said. “Within 10 minutes in my morning, playing around with the tool, I was able to do something that would typically take a staff person weeks to months to do.”

Another example Young cited was the potential use in emergency responses. In events like the recent Los Angeles wildfires, the EPA must quickly develop a communications plan, an emergency response plan, a data management plan, and a quality assurance plan, he said.

Agency staffers could use the tool to draft those plans, so emergency responders will be free “to do what they really need to do, which is responding the emergencies,” Young said.

Ultimately the EPA wants agency staff to use AI as much as possible in their day-to-day work, Young said.

“In my vision, this kind of tool and this kind of capability will be as much of our daily work as Word,” he said.

AI Skepticism

Many AI skeptics fear the technology’s potential to replace human workers—an especially heightened concern given the Trump administration’s stated desire to slash the size of the federal workforce.

For example, a 2024 report from the American Federation of Government Employees concluded that AI may “eliminate large numbers of jobs, deskill workers, erode democracy, [and] infringe rights.” Similarly, the World Economic Forum recently found that 40% of employers expect to cut their workforce because of AI.

The AFGE’s collective bargaining agreement with the EPA doesn’t explicitly address the use of AI.

The EPA has drafted a five-page document that lays out rules for agency employees, contractors, subcontractors, grantees, and volunteers in how to use any form of generalized AI.

Some of the rules include a ban on feeding sensitive or personally identifiable information into the AI tools, compliance with the EPA’s scientific integrity policy, and disclosures that a piece of work may include elements generated or aided by AI, according to the document, which was reviewed by Bloomberg Law.

Acceptable uses of generative AI include developing initial drafts of communication materials, information summarization, policy development and analysis, creativity and brainstorming, training and simulations, scientific data analysis, and environmental monitoring, the document says.

Unacceptable uses include decision making, handling of non-public information, generating misinformation, and creating biased information or altering original artifacts.

The tool is also based on an October 2023 snapshot of the Internet, so it can’t be used to do web searches, and it’s sequestered from the Internet to block external access, an EPA spokeswoman said. Every user prompt will also be logged.

EPA personnel will provide extensive training to staff to teach them how best to use AI, according to Young.

“We have a lot of younger staff that are coming in that are used to using tools like this in their day-to-day,” he said. “We’ve got a lot of middle-aged or older staff that aren’t as comfortable with these tools.”

The EPA’s action lines up with Trump’s broad support for AI, even though the project got underway during the Biden administration.

In April, the White House issued a memo calling on agencies to speed up their deployment of the technology.