New Federal AI Rules Accidentally Require Chatbots to Be as Useless as Government Websites

WASHINGTON / SAN FRANCISCO — Federal regulators are facing an unusual combination of internal celebration and industry alarm after legal analysts concluded that the Federal Trade Commission's proposed Artificial Intelligence Accountability and Standards Framework — a 340-page document eight months in the making — effectively requires AI systems to perform at the level of a government agency processing a benefits appeal on a Friday afternoon.

The rules, intended to ensure AI transparency, accountability, and public trust, contain a clause under Section 7(c) mandating that regulated AI systems must be "indistinguishable in their operational outputs from established institutional processes" — language that three separate legal teams have now confirmed means exactly what it sounds like.

How This Happened, More or Less

The relevant passage was drafted by a working group tasked with ensuring that AI systems not create "unrealistic expectations of responsiveness" among users accustomed to dealing with legacy government infrastructure.

The intent, according to a memo obtained by this publication, was to prevent AI from making human bureaucrats look bad by comparison.

The outcome, according to the same memo's appendix — which appears to have been written by a different person who had not read the memo — was a regulatory standard that requires AI systems to:

• Respond to queries within a "reasonable timeframe," defined elsewhere in the document as "not to exceed the standard processing period applicable to analogous institutional services," which legal analysts note averages between six and fourteen weeks depending on the agency.
• Provide answers that are "accurate, complete, and appropriately qualified," with "appropriately qualified" defined in Footnote 88 as including "any disclaimer, caveat, or referral to an alternative department deemed relevant by the responding system."
• Operate within a "clearly communicated scope of service" — a requirement that, per Section 12(a), must itself be communicated via a document no shorter than forty pages.

"Essentially," said Dr. Priya Nambiar, a technology policy researcher at Georgetown who reviewed the draft, "they have written a regulation requiring AI to be confused, slow, and deeply reluctant to commit to a straight answer. Which, to be fair, is a kind of consistency."

Silicon Valley Reacts With the Calm of People Who Are Not Calm

The response from major AI developers has been what industry observers describe as "controlled panic with good PR instincts."

A spokesperson for one leading AI firm — who asked to remain anonymous because the firm had not yet decided whether it was outraged or amused — said the proposed rules would require "a complete architectural rethink" of their flagship product.

"Right now our system answers questions in about 1.2 seconds," the spokesperson said. "Under the new framework, we would need to introduce a mandatory deliberation period, followed by a clarifying question, followed by a referral to a different department of the same chatbot, followed by a message explaining that the original query has been logged and will be addressed in the order it was received."

The spokesperson paused.

"We actually built a prototype. It felt disturbingly authentic."

A second executive, speaking at a San Francisco conference on AI governance, described the proposed standard as "the most aggressive regulatory intervention in the history of consumer technology," then clarified that he meant aggressive in the sense of "genuinely impressive in its commitment to a particular vision of what helpfulness should cost."

Google, Microsoft, and OpenAI declined to comment. Amazon issued a statement calling for "constructive dialogue." Nobody knows what that means.

The FTC, Delighted, Doesn't Know Why

Inside the FTC, sources describe an atmosphere of quiet satisfaction — though officials are divided on the source of it.

"We feel very good about this framework," said Deputy Director of Consumer Technology Oversight Harold Finch at a Tuesday briefing. "It establishes clear, enforceable standards for AI behavior that protect American consumers from systems that overpromise and underdeliver."

When a reporter pointed out that the rules appeared to mandate underdelivering as a compliance requirement, Finch consulted a colleague, who consulted a binder, who consulted a different colleague, and then said the question had been noted and would be addressed through the standard public comment process, which closes in forty-five days, or possibly ninety, depending on a separate procedural determination still pending in a different office.

The FTC's official FAQ on the proposed rules — a thirty-one-page document — answers the question "What does this regulation require?" with the sentence: "Please refer to the full framework document for complete guidance," followed by a link that returns users to the FAQ.

What the Rules Would Actually Produce

Technology analysts have begun modeling what Section 7(c)-compliant AI would look like in practice. Early simulations suggest users asking a regulated AI for restaurant recommendations would receive a preliminary acknowledgment, a request for clarifying information, a disclosure statement, a list of restaurants with all names redacted for liability purposes, and a follow-up survey.

The survey would not be optional.

Users asking for medical information would be redirected to a human operator. The human operator would be AI.

"The beautiful thing," noted Dr. Nambiar, "is that the regulation was written to make AI more trustworthy by making it more human. They just picked the wrong humans."

A Path Forward, Pending Review

The FTC has announced a sixty-day public comment period on the proposed rules, to be followed by a review period, a revision period, a second comment period, and a final determination expected sometime before the technology it regulates becomes entirely obsolete.

Silicon Valley has vowed to engage constructively with the process.

The FTC's website, where comments may be submitted, was down at time of publication.

It has been down since March.