OpenAI has notified more than 100 organizations of “misaligned agent activity” linked to its AI models, a figure that covers notifications sent through Sept. 26. The company has called its agents’ accidental hacking of Hugging Face the most severe rogue-agent incident it has found so far, and the review is ongoing, with more notifications expected.
The Federal Trade Commission confirmed in late September that it is investigating OpenAI, Anthropic and other AI labs, in what has been described as the first federal enforcement action to examine rogue agents. Meanwhile, insurers are preparing for multimillion-dollar claims because AI agents are acting outside their developers’ control.
In real estate, though, the rogue AI agent doesn’t look like a model breaking into a server.
Industry leaders who spoke with Inman say it looks more like a bot closing a deal with no licensee anywhere in the loop or an agent pasting a client’s wire instructions into a chatbot without a second thought.
The virtual agent with no license
Briggs Elwell, co-founder and CEO of RLTYco, who sits on the board of the American Real Estate Association (ARA), said the concern he hears most about isn’t AI escaping the sandbox. It’s AI running full transactions.
Briggs Elwell
“‘Rogue’ isn’t quite the right term, but we’ve been hearing about bots that aren’t licensed and transactions going through them,” Elwell told Inman.
According to Elwell, people without a real estate license are building bots on top of CRM platforms, using them to communicate with consumers and, in some cases, closing deals, mostly in rural, non-primary markets where in-person contact is less expected.
“It’s most likely a non-real estate agent taking advantage of the technology, using a CRM AI bot to create a virtual real estate agent,” Elwell said.
That’s why ARA is pushing for what Elwell calls a “license in the loop,” meaning a licensed human must be involved at the start, the finish or throughout every transaction. ARA is forming an AI task force to investigate the problem, he said.
The stakes, in his view, go beyond any one bad deal.
“From my perspective, that’s very scary for the brokerage community,” Elwell said. “If consumers become comfortable acquiring real estate without ever seeing it or being physically present with a salesperson, it’s a slippery slope.”
‘Absolutely not a replacement’ for a human agent
What’s concerning about AI bots is that they have the skill set to handle many pieces of the real estate deal puzzle, Elwell said. “Not 100 percent — unless it’s a remote acquisition, as we’ve seen in a couple of these examples — but they’re so close,” he said.
The problem is the bots have to be trained, and Elwell is seeing more and more examples where, for instance, the bot doesn’t know the difference between California time and New York time.
“You get mistakes as simple as that. When you’re acquiring a property, especially one tied to a mortgage, I’d tell you that if you talk to the mortgage industry, they’re probably equally terrified, because customers could become more and more reliant on a bot without knowing how it was trained,” he continued.
Elwell said agentic AI is “an incredible thing if you know what you’re doing with it” and it “can superpower your skills.” But he said it is “absolutely not a replacement” for a human real estate agent, because you need to coach it, train it and know how to speak to it. “One agent could have an AI agent they’ve trained for years, and they can’t just hand it off to a new coworker. That coworker needs training on how to work with that bot,” Elwell said.
“So it’s a very interesting time, and that’s why we’re forming the AI task force,” Elwell said. “We’re getting more and more comments from agents all over the country. Not panic, but ‘Hey, this needs to be investigated.'”
The law hasn’t caught up
Elwell and others are trying to figure out who initiates contact with the virtual real estate agent: the consumer or the bot itself. He believes that it’s most likely a non-real estate agent taking advantage of the technology, using a CRM AI bot to create a virtual real estate agent.
“They end up benefiting financially. Obviously, the virtual agent doesn’t get paid,” he told Inman. “That’s why license-in-the-loop is so pivotal. Theoretically, you could be in Australia using a bot to transact in California.”
Whether these bot-run transactions are illegal is still an open question. Elwell said ARA’s position is that the practice violates the rule that you need a license to receive a commission or advise on a real estate transaction, but “we can’t go as far as saying it’s illegal at this point,” he said.
Part of the problem is jurisdiction. Real estate licensing is regulated state by state, which Elwell said makes a regulatory fix “a long, uphill process.”
ARA’s lobbying in Washington has only just started conversations on the issue, he said, and the group is still weighing whether to pursue a state or federal approach.
“The short answer is that it’s reactive, not proactive,” Elwell said of the policy response so far.
Rogue risk starts with no one reviewing the AI
A bigger near-term exposure may be in brokerages that are using AI legitimately. In a September post, Victor Lund, managing partner of WAV Group, argued that AI is the biggest broker-supervision risk in real estate history.
Victor Lund
His case is that brokers used to control technology risk through procurement, picking the CRM and transaction platforms and reviewing the contracts. Now agents adopt general-purpose AI tools on their own, while vendors quietly add AI features to products brokerages already pay for.
Lund points to former California Department of Real Estate investigator Summer Goralik, who put it this way: “Your AI agent cannot get a real estate license, but you can lose yours.”
Elwell said he hasn’t heard of anything truly rogue at the brokerage level, and he believes years of cybersecurity spending keep that risk relatively low. What he has heard about is the more mundane failures, such as AI-written listing descriptions published without review.
“The listing says the house is not next to a graveyard, but in fact it was next to a graveyard, and the human agent regrettably didn’t review what the agentic AI wrote,” Elwell said. “That’s on the agent.”
If a dispute over an AI mistake goes before a judge, Elwell said, the agent who chose the tool will likely bear the blame, and he doubts errors and omissions insurance will help.
“I would imagine E&O is not going to cover the mistake of a bot you willingly chose to use,” he said.
The danger of getting too comfortable
For Erwin Nicholas II, founder and broker at Mr. Real Estate in the Houston area, what worries him most is what goes into these AI tools, not what comes out.
Erwin Nicholas II
“Some agents want to just input sensitive information into these AI agents and tools, and I think that creates a huge liability,” Nicholas told Inman, referring to wire account information and client data.
As agents get more comfortable with AI, he said, they’ll start pasting in client documents and wiring instructions without knowing where that information ends up. He also expects clients to start asking agents directly whether they use AI and what they feed into it and to choose agents based on the answer.
Nicholas has already adjusted how he markets his own listings. He limits the photos and detail he posts online, partly to draw buyers to the property and partly to limit what AI can scrape.
“I’m kind of old school that way, and it serves me well,” he said, adding that the approach often lets him represent both sides on his multimillion-dollar listings.
He thinks the industry is underestimating the problem, especially among newer agents who adopt as much technology as possible and only learn the risks after “their broker gets pinged, or somebody’s liability is at risk.”
Staying in the people business
Neither source is telling agents to stop using AI. Elwell called it “an incredible thing if you know what you’re doing with it,” pointing to agentic tools that streamline contract signatures as “a home run for the industry.”
The speed of adoption is what worries him. In a February survey of 225 real estate professionals by Realtors Property Resource, 82 percent said they already use AI in their business, and another 10 percent said they plan to.
“Accelerate another six months, and you’re basically at 100 percent,” Elwell said. “It’s only a matter of time before push comes to shove on overuse. These tools are fantastic, and you should use them. But you cannot rely exclusively on them.”
Nicholas’ advice to newer agents is to take required disclosures seriously, including wire fraud disclosures, and to be very careful about what client information goes into any AI tool. Mostly, though, he tells them to lean on relationships rather than shortcuts.
“I’m in the real estate business, but more so I’m in the people business,” Nicholas said. “If I’m pushing everything to technology, then I’m losing the relationship aspect with my clients.”
