Reflections from Houston: AI, MLS Operations, and the Policy Gap
On Thursday at NAR NXT, I served as a guest speaker in a 90-minute session with a room full of MLS and Association executives, sponsored by MRED, for a candid discussion about the excitement and risks associated with Artificial Intelligence.
The range of experience in the room was wide. Some had never interacted with any model at all. Others had tinkered. A few execs and their team are already working on drafting organizational AI policies. That alone tells you where the MLS world is right now. We have scattered levels of understanding, high interest, and almost no shared framework to guide org-level decisions.
My role was to offer a broad MLS perspective on how MLSs are beginning to approach AI from both operational and licensing angles.
One strong consensus emerged quickly. MLSs are largely on their own here. NAR is not expected to provide sweeping guidance, which means MLSs must define their own language around use cases, limitations, allowances, member expectations, and rules governing how we construct our policies in an AI context. This includes clarifying acceptable use, prohibited use, and the practical realities of model training. And that’s just a starting point.
A major concern raised in the room was the widening compliance gap. Case in point, large language models are not designed to clearly acknowledge when they lack information. Imagine the moment when a consumer asks a model about home value and receives a wildly inaccurate answer. Who bears the responsibility? This simple scenario remains unresolved and it will become a central policy issue sooner rather than later.
The conversation also echoed an earlier, pivotal moment in MLS history. In the early 2000s, when MLS data began appearing on new and unregulated websites, the industry had no policy framework. IDX quickly emerged, bringing structure to the chaos and stabilizing the ecosystem for the next 25 years. But the challenge AI presents is far more complex. Checking a website for compliance is straightforward. Auditing an AI model is not. Our oversight tools are still rooted in that earlier era while the technology we are attempting to regulate has moved far beyond it.
This Houston meeting (thanks again Rebecca and Chris) produced a number of insights and a long list of questions that still need answers. It reinforced that MLSs cannot afford to wait for direction. This is the moment to study how other industries are approaching AI governance, listen to people with practical experience, learn how models actually work, and build policies that anticipate rather than react.
AI is already improving operations for those who embrace it, but it also deserves scrutiny, structure, and clear boundaries around how MLS data is handled. The organizations that begin this work now will be the ones best positioned for what comes next.
