As the Federal Communications Commission moved to ban Chinese laboratories from testing smartphones and other consumer electronics intended for U.S. consumers, it sought input from the public on an initiative with national security implications.
But federal officials may never actually read the comments, government records obtained by Straight Arrow show. Instead, the officials fed the comments into an artificial intelligence chatbot that boiled down experts’ policy analyses into succinct summaries.
Download the Straight Arrow app today to get the stories that matter free from manipulation, bias or agenda.™
Point phone camera here
The records, obtained through a Freedom of Information Act request, reveal a broader FCC practice of using commercial chatbots, including Anthropic’s Claude, to streamline the rulemaking process. Another AI summary included in the records is about national “drone dominance.”
Federal laws require the FCC and other government agencies to consider public feedback before adopting new rules, including those pertaining to national security. Whether anyone in the government really reads the public comments is another question entirely.
As the Trump administration goes all-in on AI across the federal government, the tools have been framed as a way to improve efficiency. But they’ve also raised concerns about accuracy, algorithmic bias and accountability. In this case, the government’s use of AI to process public comments highlights questions about whether people are awarded a full opportunity to weigh in.
“This is really about the fundamental right to participate in your government, and we shouldn’t just be summarizing that,” said Katie Tracy, the senior regulatory policy advocate at Public Citizen, a left-leaning advocacy group.
Inside a government ‘comment summarizer’
In a memorandum last year, the White House Office of Management and Budget called on agencies across the federal government to expand their AI use with a focus on “innovation, governance and public trust.”
The FCC discloses the chatbot tool in its AI use cases inventory, published in February, which noted at the time the technology was in “development or acquisition status.”
The use case inventory notes that manually reviewing and analyzing public comments is “time-consuming, inconsistent and resource-intensive.” Automating the “identification of key legal themes, sentiment and duplicate content,” the commission notes, allows for “faster, more consistent and scalable legal analysis.”
The documents obtained by Straight Arrow offer specific insight into how federal officials are using AI to guide the rulemaking process, which some advocates warn could have ill-intended consequences.
Specifically, records show, commission staff are running public feedback through an “FCC Comment Summarizer” that relies on two AI agents: Claude Opus 4.8 by Anthropic and Inkling by Thinking Machines Labs. The tool analyzes, categorizes and summarizes public comments.
The FCC didn’t respond to emails from Straight Arrow seeking comment.
In an email in July, Alice Suh Jou, the chief of staff and deputy chief of the FCC’s Office of Engineering and Technology, wrote that she “created an AI summary” of public comments “responding to Bad Labs 2.0,” an FCC effort to ban electronics vendors from hiring “foreign advisory-owned” product testing labs, specifically those based in China.
The FCC voted unanimously in April to advance a proposal that would ban all Chinese labs from testing smartphones, computers and other electronics intended for the U.S. market, claiming national security concerns. About three-quarters of all electronics sold in the U.S. are tested in China, according to the commission.
In total, the AI summarizer boiled down 81 public comments into brief snippets — complete with sentence structures that have become telltale signs of AI-written text.
“Commenters are sharply divided over whether to prohibit and withdraw recognition of labs in non-Reciprocal Economies,” one AI summary notes. While national security advocates “strongly back the prohibition” of the testing labs, according to the summary, opponents including foreign governments, international manufacturers and trade associations “argue the proposal improperly conflates security risk with reciprocity.”
In one summary on the “costs, benefits and consequences” of the proposal, feedback from 66 organizations, including the National Association of Manufacturers and the Foundation for Defense of Democracies, was boiled down to 129 words.
FCC officials appeared to rely on the summaries, though the degree to which they inform government regulations remains unclear.
“This is helpful,” Katherine Nevitt, an FCC senior attorney, responded.
‘Extremely alarming’
The FCC isn’t the only federal agency leaning in on AI to guide policymaking. At the Department of Housing and Urban Development, for example, officials are leveraging AI in a bid to better understand how the recipients of some $77 billion in grants spend the money.
That initiative, while designed to detect and prevent fraud, has led to concerns of AI errors and the potential to be used against political rivals.
Efforts to inject AI into the federal rulemaking process have also raised legal questions. While the federal rulemaking process seeks to improve government policies, courts have weighed the procedural requirements “in a pre-ChatGPT world” with “human intelligence in mind,” attorney Tara Aida wrote in a recent article for the Yale Journal on Regulation.
“If agencies over-rely on AI to carry out these procedural tasks,” Aida wrote, “they threaten to undermine” the goal of improving final rules. It could also open agencies to litigation, Tracy said, because they’re required to base rules in reasoned decisionmaking.
Although AI chatbot responses can be convincing, ultimately they’re simply “statistical predictions of which words are likely to appear together,” according to a recent Public Citizen blog post. Ultimately, the outputs are “not grounded in subject-matter expertise, institutional knowledge or the judgment of career staff.”
While chatbot prompts are prone to producing inaccuracies and biases — including hallucinating fabricated information — federal regulations are highly technical. Overreliance on AI in the rulemaking process, according to the post, could give “federal workers unwarranted confidence in inaccurate and false information.”
Tracy told Straight Arrow that the government’s embrace of AI comes at a time when even technology insiders are warning of potentially dire consequences. Former Anthropic AI researcher Jacob Coxon announced in a viral social media post this month he was resigning from the company due to concerns that it wasn’t acting responsibly. As Anthropic and other AI companies compete to build “self-improving superintelligence,” he wrote, the companies are “gambling with our lives.”
Coxon’s warnings, Tracy said, should give government agencies pause.
“We have this unregulated product that people are claiming is incredibly harmful for humanity as it exists, and we’re still using it to create really important public decisions in agency rulemaking that affect people’s lives,” Tracy said. “It’s extremely alarming.”
The stakes are particularly high when the AI summaries pertain to proposed rules related to national security, Tracy told Straight Arrow.
Among them, the public records show, is a Trump administration effort to ensure U.S.-based companies have the resources they need to “develop a domestic drone ecosystem for commercial and military applications.”
“Thank you again, in advance,” Tom Struble, an FCC legal adviser, wrote in an email in May, “for creating another AI summary of these drone comments!”
Round out your reading
- Lawyer says 57 Cornell frat members were in the group chat. Could their silence be a crime?
- Florida county discovers mysterious Flock cameras with no obvious owner.
- Off Script: They won their primaries on one message. Now it’s disappearing.
- Some question whether a popular affordable housing policy is making the nation’s home shortage worse.
- Kelvin wave incoming: Why the US should care.
