The White House's new AI chatbot promised Americans they could ask 'any question' and receive an up-to-date answer. Yet just hours after launch, it began refusing to answer questions about the Epstein files.
The awkward U-turn came after America.gov initially answered politically sensitive questions — including questions about whether Donald Trump's name appeared in the Epstein files — before suddenly responding that it could not answer political questions. The change was reported by the Associated Press, which found that the chatbot had modified some of its answers after launch.
That raises the question at the heart of this episode: can a government AI genuinely be called an information tool if its answers on sensitive subjects can change almost overnight?
The White House's own description of America.gov sets a high bar. When the website launched on 29 September, the administration said Americans could ask any question and receive an up-to-date answer. It presented the platform as a single digital point of entry for government information and services.
Information gains trust through consistency, not just accuracy. But when an AI system gives one answer today and refuses to discuss the same subject tomorrow, users are left wondering what changed — the information, the system, or the rules governing it.
And that is where the issue becomes bigger than one question about Epstein.
The concern is also about the extent of government control over the information shared with the public. One day, the public can access an answer; the next, that access is denied. If a chatbot is being positioned as a convenient gateway to official information, who decides which questions it can answer?
It is important, however, not to jump from the chatbot's changing responses to the conclusion that the White House personally ordered the system to suppress questions about Epstein. The AP reported that the answers changed and that the chatbot began refusing certain political questions, but that reporting does not by itself establish who made the decision or why.
Questions of AI Models Being Used in Government Systems
AI systems can change because of technical updates, new instructions, safety rules, changes to their underlying models, or adjustments to the information they are allowed to access. Without clear explanations, however, users may struggle to understand which of those factors is responsible.
So, can a government AI be called an information tool if its answers on sensitive questions can flip almost overnight?
That uncertainty matters because AI is increasingly being positioned as a shortcut to information. Instead of searching through government websites, public documents and official statements, users are being encouraged to simply ask a chatbot and trust the answer it produces.
The White House has previously said federal AI systems should prioritise truthfulness, historical accuracy, scientific inquiry and objectivity, while also being neutral and nonpartisan.
That makes transparency even more important.
If a government chatbot is going to become a front door to official information, Americans need to know why an answer changes, what sources the system relies on and where its limits are. They also need to know whether an unanswered question reflects a lack of information, a technical limitation, or a deliberate restriction.
Otherwise, the biggest question surrounding government AI may not be 'What does it know?' — but 'Who decides what it is allowed to tell us?'
