On Tuesday, the Trump administration launched America.gov, its artificial intelligence chatbot, to be the new way Americans get answers to their most pressing government-related questions. For critics, the chatbot’s answers only create more questions about users’ security.
America.gov is already a punchline for Democrats to share how the chatbot’s answers to questions radically contradict President Donald Trump’s own past statements about who won the 2020 election. To the Trump administration, this chatbot is the future of how Americans interact with the government.
Trump’s executive order has grand ambitions for the chatbot, stating that America.gov will not only be a search engine for government information but “the single point of entry for Americans to access covered services online.” Americans will be able to use the chatbot to renew passports and to enroll in Medicare later this year.
If users ask for help with questions related to their Social Security number or Medicare number, America.gov’s chatbot will caution against sharing that specific information. The chatbot’s website also states that “messages aren’t saved by America.gov, and your chat disappears when you leave. Some answers may be cached for up to two hours.”
Despite these disclaimers, the type of sensitive and identifiable information people could be sharing in these chat logs concerns security researchers.
Nicole Alvarez, a senior technology policy analyst at liberal think tank Center for American Progress, said your Social Security number is an obvious identifier for the chatbot to screen out, but there could also be equally sensitive details that could reveal who you are.
For example, someone might share, “‘I’m a 70-year-old veteran living in Arizona, and I recently got a cancer diagnosis. I got laid off, so I lost my employer insurance last month, and I need help figuring out what the benefits are for me as a veteran. How can I apply for Medicare? What will cover my treatment?’” she noted. “Disclosure of [this information] or misuse of it could expose people to discrimination within the same government [or] fraud, depending on who has access to that information and targeting.”
Trump has acknowledged that America.gov could face a hacking threat, but that the U.S. government can “counteract” high-IQ hackers.
“I would absolutely advise against submitting sensitive information or using it to apply for any benefits when those features are added,” said Philip Fields, a cybersecurity researcher and former FBI counterintelligence analyst. For him, the risks of wrong answers are too high and Trump’s government has a track record of not securing Americans’ privacy.
When HuffPost asked the White House about critiques of the new site’s privacy, a White House official responded: “These researchers are idiots. No data or messages will be stored by America.gov, nor will they be shared with Gemini or Grok for LLM [large language model] training,” which are the two AI models that power America.gov.
“It’s not fair to call people idiots if they have suspicions around how this particular administration is handling their information, specifically their sensitive government information,” Alvarez said in response. “Honestly, I think you’d be an idiot to not question it.”
In response to the White House, Fields cited America.gov’s own privacy policy, which states that America.gov will “retain Cloudflare security logs, which may include your IP address, for 12 months.”
“They’re not storing queries, but they are storing information about visits. That’s still data,” Fields said. “Your IP address and the time that you visit a website shows your behavior as a user on the internet.”
Security Researchers Caution Against Sharing Personal Info With America.Gov
Kevin Dietsch via Getty Images
Security researchers cited past incidents of reported misuse of Americans’ data under the Trump administration as a reason for their caution of the chatbot.
Fields cited a June Guardian investigation, which found that the National Design Studio (which designed America.gov) installed visitor-tracking software on federal websites despite privacy laws that require public disclosure of the collection of personal information. (This tracking software got removed after the Guardian investigation and the White House told the outlet that its “personnel comply with all legal requirements.”)
“My initial concern is that this team is already known for abusing public trust and not following proper IT security protocols,” Fields said.
Alvarez said the Trump administration has been accused of reusing Americans’ sensitive personal information in one federal agency for an entirely different purpose without the individual’s consent. In Alvarez’s report in Center for American Progress, she cites the IRS, which reportedly had shared taxpayer data with Immigration and Customs Enforcement for the purpose of finding and deporting tax-paying individuals, as one example of this kind of “secondary data abuse.” This alleged secondary use has been challenged and blocked in litigation.
Consider what technology powers the chatbot as well. OpenAI’s ChatGPT, xAI’s Grok, or Google’s Gemini have all been found to “hallucinate” or make up certain information in answers. In general, people should at least “double-check” and verify the answer if they’re using an LLM, Alvarez said. “Make sure that what it’s saying is verified because we are seeing inconsistent things come from this administration,” she said.
The risks of receiving incorrect or incomplete information using America.gov is not worth it to Fields, either.
“When you apply for government benefits, you are responsible for making sure your application is true and correct,” Fields said. “And if this thing messes up … that’s a risk I wouldn’t take.” His advice would be to use an agency’s direct phone helpline or website over asking the chatbot.
“Especially when it comes to services like applying for Medicare, something that is personal and so necessary, it’s best to not just rely on this,” Alvarez continued. “And this goes for any chatbot.”
This article has been updated to clarify Alvarez’s views on LLMs and allegations on the Trump administration’s reported use of sensitive data.
