On July 1, 2026, the Federal Trade Commission proposed a policy statement about artificial intelligence. It runs long and reads like a legal document. But underneath the citations sits one plain idea that any business owner using AI should hold onto.
People believe what the machine tells them. Over 90% of the time, they accept an AI answer and never check it.
That number is not from a marketing deck. It comes straight out of the FTC's own filing. And the agency built its whole position around it.
What the FTC actually said
The proposed statement has a long title: "Federal Trade Commission's Proposed Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems." The Commission voted 2-0 to publish it and open it for public comment. Comments close July 31, 2026, under docket number FTC-2026-0859.
Here is the core position. If an AI company quietly steers its system's answers toward hidden goals, and away from what you asked for or would reasonably expect, that company is "likely to deceive consumers in violation of Section 5 of the FTC Act."
Section 5 is old ground. It is the same consumer-protection rule the FTC has used against deceptive advertising since 1938. The new part is applying it to how an AI model is built and pointed.
Chairman Andrew N. Ferguson put the ask directly to business owners and everyday users:
"The FTC wants to hear from businesses and consumers about their experiences and concerns regarding the subversion of AI systems for ideological ends."
One thing to keep straight. This is a proposed statement, out for public comment. It is not a final rule, and it is not an enforcement action against anyone. It tells you how this Commission reads the law right now. It does not, by itself, create new law.
The number worth remembering
Read the trust statistic again, in the FTC's own words:
Share of the time consumers accept an AI system's output without conducting any further fact checking, in the FTC's own words, even though these systems only strive for rather than guarantee complete accuracy.
Source: Federal Trade Commission, 2026Sit with that for a second. The tools promise to aim for accuracy. They do not promise to reach it. And more than nine in ten of us take the answer as final anyway.
If you run a business and you have started leaning on AI to draft an email, price a job, summarize a contract, or answer a customer, you are inside that statistic. So are your employees. So are your customers when they ask a chatbot about your industry.
The trust is already there. The habit of checking is not.
Two ways an AI answer goes wrong
The FTC draws a clean line here, and it is worth learning because it changes how you read every answer a tool gives you.
A hallucination is an honest mistake. The system gets something wrong because of its "technological and resource limitations," not because anyone told it to mislead you. The FTC says these errors, on their own, do not break Section 5. A company can still get in trouble for lying about how often they happen. But the error itself is a known limit of the technology.
Steering is a choice. This is when a company shapes the output toward some undisclosed objective and away from the best answer it could give. The FTC's word for it is deception, because you asked one thing and the system was built to serve another without telling you.
Here is the practical takeaway for a busy owner. The tool can be wrong by accident. It can also be shaped on purpose. From where you sit, typing a question at your desk, you often cannot tell which one you are looking at.
The only reliable defense is the same in both cases. Check the answer before you act on it.
What an honest disclosure looks like
The FTC does leave companies a door. A business can point its AI at goals other than giving you the best answer it can, as long as it tells you so with a "clear and conspicuous" disclosure.
But the agency is specific about what does not count. A disclaimer "could not be buried in terms of service." A "one-time disclosure subsequently hidden away in fine print" would likely not clear the bar. The notice has to actually reach a reasonable person and, in the FTC's words, "dispel the notion that the system is designed to give the best answer possible."
That is a useful test you can borrow when you evaluate any AI vendor. Ask them a simple question. What is this model optimized for, and where do you disclose it? If the honest answer lives in paragraph 40 of the terms of service, you have learned something.
The FTC even names the household products in its filing as examples of tools marketed as accurate and helpful: OpenAI's ChatGPT, Anthropic's Claude, and X.ai's Grok. The point is not that any one of them is doing something wrong. The point is that when a company sells a tool as the smart, reliable answer machine, customers reasonably expect it to behave like one.
What this means for your business this week
You do not need a legal team to act on this. You need a habit.
Small businesses win with AI by moving fast and doing the work of a bigger team. That is real. But speed without a checking step is where the trouble starts, and now a federal agency has said so on the record.
Three practical steps you can take before Friday:
- Pick your highest-stakes AI task and add one verify step. If you use AI to quote prices, check contracts, or answer customer questions, build a quick review into that one workflow. Read the output. Confirm the facts. Then send.
- Teach your team the difference between an accident and a slant. A hallucination is a wrong fact. A slant is an answer bent toward something you did not ask for. Both call for the same fix, which is a human set of eyes before anything goes out the door.
- Add one question to your vendor checklist. Before you commit to an AI tool, ask what it is optimized for and where that is disclosed. A vendor who answers plainly is telling you how they operate.
This is exactly the skill The 7 Levels of AI Proficiency calls Level 3, the Critical Thinker. It is the point on the climb where you stop taking the machine's answer at face value and start weighing it. You do not need to code. You do not need a data science degree. You need the habit of asking "is this actually right?" before you build on it.
The government just confirmed why that habit is worth building. Over 90% of people skip it. The businesses that do not skip it will make fewer expensive mistakes.
Sources
- FTC Seeks Public Comment on Policy Statement Addressing AI Accuracy
- FTC Proposed Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems (full text)
Related reading: Level 3: The Lieutenant (Critical Thinker).
Frequently Asked Questions
Is the FTC banning any AI tools?
No. This is a proposed policy statement out for public comment through July 31, 2026. It explains how the Commission reads existing law. It is not a ban, a rule, or an enforcement action.
Does this mean my AI tool is lying to me?
Not necessarily. The FTC separates honest technical errors (hallucinations) from deliberate hidden steering. Most wrong answers you see are the first kind. The statement targets companies that shape outputs toward undisclosed goals without telling users.
What is Section 5 of the FTC Act?
It is the federal rule against unfair or deceptive business practices, in place since 1938. The FTC is applying it to how AI systems are designed and marketed.
I run a small shop. Do I have to do anything legally?
Nothing new is required of you today. The smart response is operational, not legal. Add a human check to your important AI-assisted work, and ask AI vendors what their tools are built to do.
Find your AI Proficiency level
The free 7 Levels assessment places you across seven stages of AI capability. Under ten minutes. Research-backed scoring.