AI Governance

You Named Your Chatbot. A Regulator Just Wrote Down Why That Matters.

Every management company in the United States should read this and prepare accordingly. Not because Colorado law reaches your communities. Because the question Colorado just answered in writing is the same question Maine, Utah, California, and New York are already answering in their own ways.

Colorado published proposed AI rules on August 11. One paragraph lists the evidence a state uses to decide whether a chatbot has stopped being a tool. The first item on the list is a name.

Chatbot, assistant, agent, concierge, virtual leasing specialist. The label is not the test. Colorado regulates a conversational artificial intelligence service, and what matters is whether the thing talks to a consumer. If yours does, this is about yours.

Multifamily gave its chatbots names. We gave them genders. We gave them avatars, personalities, and in more than a few cases, backstories. We put them in job postings. We introduced them to residents as though they were members of the on-site team. We did all of it because it tested well.

On August 11, 2026, the Colorado Attorney General filed proposed rules implementing two new state AI laws. Buried in Rule 8.2 is a list of factors the Department of Law will weigh when deciding whether a conversational AI service is designed to simulate emotional companionship.

The factors include whether the service:

• Includes anthropomorphic design features such as personalized dialogue and the ability to assign a chatbot a name, gender, avatar, or fictional backstory
• Was trained, produced, and deployed in a way that prioritized, promoted, or failed to address sycophantic behavior, or outputs that simulate empathy and offer users nonjudgmental responses and continual validation
• Has the ability to recall and respond to a user's characteristics, preferences, and past conversations

Read that second factor again. A government agency wrote down that building a bot to be agreeable, validating, and never confrontational is evidence of something. That is not a fringe design pattern in this industry. That is the design brief we handed to vendors.

Be careful about scope, because the scope matters

I care more about being right than about being alarming, so here is the honest version.

Those factors do not float free. They attach to two specific exemptions in Colorado's Chatbot Safety Act, House Bill 26-1263. One covers educational tools used in a school setting. The other covers conversational features embedded inside a larger application that is not itself a chatbot. Both are conditioned on the service not being designed to simulate emotional companionship or encourage emotionally dependent interaction. Those are the only two exemptions in the statute that carry that condition, and Rule 8.2 cites exactly those two.

There is a separate exemption, and it is broader. Section 6-1-1701(3.5)(b)(II) excludes a service primarily designed to provide commerce-related or transactional assistance, including product or service recommendations, shopping, ordering, payments, delivery, returns, customer support, or customer service. Most leasing bots likely fall under it.

Your leasing bot is probably not regulated by Colorado's Chatbot Safety Act. Confirm that with your own counsel, not with me and not with a blog post.

Now here is why it still matters, and why it matters outside Colorado.

The exemption is not unconditional

Read the words of that commerce exemption carefully. It protects a service primarily designed to provide commerce-related or transactional assistance.

The phrase doing the work is "primarily designed." That is a question about what the thing was built to be, and it is answered by the design record, not by the label a vendor puts on the invoice.

A bot built to check availability, answer a pricing question, and schedule a tour sits comfortably inside that exemption. But every hour spent on the name, the face, the backstory, and the memory of past conversations is an hour spent on something that serves a relationship rather than a transaction. Nobody gives a shopping cart a childhood. The more of that work there is, the less accurate "primarily designed for transactional assistance" becomes as a description of what was actually built.

To be clear, that is an argument, not a prediction. The Attorney General wrote clarifying factors for four of the twelve exemptions and left the commerce exemption alone entirely. Nobody knows yet how that phrase will be read. I raise it because the risk is asymmetric, and because it is the question a smart opposing counsel asks first.

If you do not operate in Colorado, read this part twice

Colorado is the loudest example, not the only one. Several states already regulate chatbot disclosure, and they do not share a trigger. That difference is the whole point.

Maine enacted the Chatbot Disclosure Act in June 2025, effective September 24, 2025, codified at 10 MRSA section 1500-DD. It requires a business using a chatbot to communicate with consumers to notify them that they are not interacting with a human in cases where a reasonable consumer could not tell the difference. It is enforced under the Maine Unfair Trade Practices Act. There is no company-size threshold and no intent element.

Sit with that trigger for a moment. The obligation attaches when a reasonable consumer could not tell the difference. Naming your bot after a person, giving it a face, and writing it a personality is the single most effective method available for making a reasonable consumer unable to tell the difference. The design choice does not merely risk running afoul of the standard. It manufactures the condition that switches the standard on.

Utah requires a truthful answer when a consumer asks directly whether they are dealing with AI, under its AI Policy Act effective May 1, 2024. California enacted SB 243, effective January 1, 2026, covering companion chatbots, with a private right of action. Customer service bots are currently excluded, though a pending bill would extend disclosure duties to them. New York enacted a companion chatbot statute effective November 5, 2025. And the Federal Trade Commission Act reaches deceptive practices nationwide regardless of whether a state has passed a chatbot statute at all.

The specific requirements will keep moving. The direction will not. Every one of these regimes turns on the same question: did the person know what they were talking to. A human name is a thumb on that scale, in every state, under every trigger, forever.

Where Fair Housing actually enters

In May 2024, HUD issued two guidance documents. One addresses the screening of applicants for rental housing. The other addresses the advertising of housing and other real estate-related transactions through digital platforms. Both rest on the same principle: the Fair Housing Act applies to housing decisions regardless of who makes them or what technology is used, and a housing provider cannot transfer its statutory obligations to a vendor, an algorithm, or a scoring model. Housing providers and the companies that build these tools each carry independent obligations. A vendor's compliance assurance does not insulate the provider.

Be precise about what that does and does not cover. Neither document is about conversational chatbots as a category. A bot that answers a question about office hours falls under neither.

But a bot that asks about income, criminal history, rental history, or source of income is performing screening work. A bot that describes some communities to some prospects and different communities to others is performing advertising work. Which category a tool lands in is determined by what it does, not by what it is called.

Here is where the name comes back. When a complaint is filed, the question is what the prospect was told and what they reasonably understood. A statement delivered by "Ava from the leasing team" and a statement delivered by "Community Leasing Assistant, an automated tool" are the same words carrying different weight. One reads as a representation from a person with authority. The other reads as machine output. You chose which one you built.

What to do this quarter

None of this requires waiting for a final rule.

1. Inventory every conversational AI touching a prospect or resident. Website chat, SMS, voice, email automation, application follow-up. Most organizations cannot produce this list on demand, which is itself the finding.
2. Write down what each one is named and why. If the answer is that it tested better, write that down too. That sentence is the record.
3. Sort each tool by what it does, not what it is called. Does it screen, advertise, or answer. Screening and advertising carry Fair Housing exposure the moment an automated system touches them.
4. Check the disclosure against the strictest state you operate in. Not the average. The strictest. A banner satisfies some regimes and none of the others, several of which require runtime behavior rather than static notice.
5. Ask your vendor what the bot remembers. Whether it retains characteristics, preferences, and past conversations, and whether that retention is configurable. Get the answer in writing.
6. Ask whether it was tuned for agreeableness. If the vendor optimized for warmth and validation, you want to know before someone else asks.
7. Name your bots what they are. Not a person. A tool. Naming a system after its function costs nothing and removes an entire category of question you would otherwise have to answer later.

Name your bots what they are

AI is infrastructure. It is not a mascot.

We are past the point where warmth is a free choice. Every anthropomorphic decision made for conversion is a decision that now sits in a design record, and design records get read by people who were not in the room when the decision was made. Build the tool. Disclose the tool. Call it a tool.