Real estate agents and brokers are using AI every day, freeing up their time to focus on building relationships and getting more face time with clients. But popping their favorite chatbot’s extension onto their browser or letting it download MLS data has a major downside: It’s putting your client’s data at major risk.
If that information falls into the wrong hands, it can do lasting damage and put brokers in a tough legal spot. And at least one state is already saying that brokers are responsible for advice given out by AI chatbots, who are essentially engaging in activities that require a license.
As the industry’s adoption of AI soars, some MLSs aren’t taking a backseat and waiting for federal regulators and real estate associations to figure out solutions. Instead, they’re setting up their own limits to protect the integrity of their data—and their organizations.
Proactive approaches emerge
Rajeev Sajja, chief AI and product officer with Bright MLS—one of the largest MLSs in the U.S. with more than 100,000 subscribers—said the organization is already in talks with vendors that want to train AI models on its listing data. Bright MLS has built a licensing structure that treats AI training as a separate addendum to its standard data agreements, so anyone wanting that level of access has to ask for it explicitly, he added.
By mid-September, Bright MLS will roll out a Model Context Protocol (MCP) server that lets subscribers plug an AI assistant directly into Bright’s data instead of downloading raw files or pasting information into external tools. This will help eliminate the guesswork for brokers and agents as to whether they’re handling MLS data compliantly when they use AI.
“The world we’re all headed toward is an (Application Programming Interface) type of metering where you can’t really download the data; you call the data when you need it, and we can track and meter usage,” Sajja said, adding that smaller and mid-sized MLSs will likely lean on vendor-built versions of the same approach as a “foundation for governance.”
Controlling data flow is trickier than it seems
Data governance is the issue keeping Bright and other MLS leaders on edge. Sajja confirmed that Bright discovered major AI platforms were surfacing copyrighted listing photos and descriptions from its listings without permission. While Bright was able to get the AI companies to stop (for now), he’s not counting on voluntary compliance in the future.
“I don’t think we can trust the large AI companies to follow guardrails unless you have proper guardrails,” Sajja said, adding that “plead for forgiveness versus permission seems to be the approach” when it comes to potential violations.
Art Carter, president and CEO of California Regional MLS (CRMLS), isn’t keen on letting AI companies police themselves. Carter said he tested the boundaries of how chatbots handle MLS data recently by asking Claude how to obtain CRMLS’ data for a hypothetical product. The chatbot offered two paths: license it directly, or install a Chrome browser extension that would log into a subscriber’s account and pull whatever was needed.
Both approaches are problematic, Carter said. The unauthorized plug-in becomes an issue for the vendor responsible for the organization’s back-end systems, and this type of non-compliant exposure happens without anyone at the MLS ever knowing.
CRMLS’ answer is NexusRE, a governance layer WAV Group built for REcore, a company Carter also leads. NexusRE sits between an MCP server and the underlying MLS database, allowing brokers and MLSs to see how AI platforms access, use and monetize listing data. It applies permission-based rules to every user query in real time, acting as “one door” for data to legally exit the MLS, Carter said.
Agents, brokers may not realize they’re compromising data
For Victor Lund, principal of WAV Group, the more urgent problem is the everyday habits of agents who have no clue of the dangers their AI usage poses to MLS listing data.
Adding a chatbot’s browser extension gives the AI tool the ability to see every page an agent views, including confidential MLS listings. And it happens without any credentials being stolen or systems being breached, Lund said.
“The exposure is invisible because it looks completely like normal behavior,” Lund said.
Another issue is copying MLS data and pasting it into a chatbot, especially into a free account. While paid subscriptions allow users to opt out of storing chats and letting the tool use their inputs to train large-language models, or LLMs, free versions typically don’t offer that option, Lund explained.
The financial and legal risks tied to AI scraping of listing data are enormous.
Take listing photos, which are copyrighted works; federal statutory damages start at $750 per violation, Carter said. With millions of photos in the CRMLS database alone, he said a worst-case claim tied to unauthorized AI usage means an AI company could be hit with hundreds of millions of dollars in damages.
Carter is also worried about the accuracy of AI-generated real estate advice offered to consumers. Because AI models make inferences to fill gaps in incomplete data, he said, wrong information from a chatbot could lead buyers and sellers to make costly decisions.
Another prickly issue is who’s liable when that happens. Carter pointed out that brokerage AI tools that answer real estate questions are effectively engaging in licensed activity without a license. California’s Department of Real Estate has made it clear to Carter that the brokerage is responsible, because the AI tool is considered an unlicensed assistant under their control.
“Harmed individuals are going to find somebody to sue, and I just think brokerages and agents need to be very careful of that concept,” he said.
The path forward for data safety
While some MLSs have policies that discourage uploading MLS data into free AI services, there’s no current way to enforce it.
Bright’s forthcoming MCP rollout, along with similar solutions from CRMLS and vendors including FBS and Cotality, are meant to close the data loopholes that could allow AI scraping and mass downloads. Some MLSs are going as far as adding explicit AI-training bans directly into their data licensing agreements.
However, these policies are not a real deterrent, Lund cautioned.
“You can’t ban your way out of this,” he said. “Nobody can enforce it.”
Lund added that MLSs should focus on technical fixes, such as marking web pages as unindexable to AI crawlers, or embedding ownership metadata into listing photos that can’t be stripped out or edited away. For example, Cotality’s Trestle Defender makes broker attribution part of the photo file itself, which goes further than traditional image watermarking, Lund said.
Meanwhile, Sajja maintains that MLSs are “the most comprehensive resource for data,” but he wants to take that a step further and ensure that they also “become the trusted intelligence layer in the AI era.”
“There needs to be an AI component because I think the only way you innovate is if you can train on the MLS data to serve up things in an experience…where the subscriber is,” Sajja said.
As the industry continues to embrace AI at scale, inaction on MLS data safety simply isn’t an option anymore. The time to act is now, Carter said.
“AI is going to come up and slap people in the face pretty hard in a very short period of time,” Carter said. “They’re busy worrying about these private listing networks and all of this policy stuff that…in a year won’t matter. We are fighting over deck chairs on the Titanic.”
