Legislators want to know how the Securities and Exchange Commission is overseeing agentic artificial intelligence trading on registered brokerage platforms.
In a letter from Democratic members of the House of Representatives’ Financial Services Committee to SEC Chair Paul Atkins this week, lawmakers were particularly worried about customers’ use of third-party AI agents in SEC-registered brokerage platforms. In the letter, legislators (including U.S. Reps. Bill Foster and Brad Sherman, the respective ranking members of the Financial Services and Capital Markets Subcommittees) noted that brokerage platform disclosures clarify that their platforms can’t guarantee the accuracy of agents’ outputs and that agentic trading involves “significant risk.”
How, they asked Atkins, does characterizing AI agents as third-party tools align with retail investors’ expectations for investor protections on the platforms?
“The AI firms developing and deploying these agents have thus far operated largely outside the securities regulatory framework, even though their systems are making or enabling consequential investment decisions on behalf of retail investors,” the lawmakers wrote.
While brokerage firms continue to develop internal AI agents to boost operational efficiency (and agentic trading assistants for investors), some platforms are welcoming third-party agents.
Late last month, Robinhood Financial declared it was “open to agents,” allowing customers to build them and offering direct access to Robinhood’s platform, “without the workarounds of unofficial APIs holding you back elsewhere.” The firm also unveiled its “Agentic Credit Card,” a system that allows agents to spend on clients’ behalf.
According to the Democratic lawmakers, if agents are trained on similar data, there’s a risk of “correlated trading decisions” and a form of “herding behavior” that could increase volatility or heighten market stress (and the risks only heighten with AI agents’ ability to trade rapidly and at scale).
In response, the lawmakers asked Atkins to answer 13 questions, probing the extent to which the agency has worked with retail brokerage platforms on their allowances for agentic trading (and whether the agency offered any no-action relief or formal approvals to platforms or AI developers).
Additionally, lawmakers want to know what responsibility the agency believes AI developers bear when their agents execute trades on behalf of clients, and whether the use of a third-party agent on a platform would “alter, limit or absolve the broker/dealer of any obligations as registered broker/dealers under the federal securities laws.”
“Will AI agents that facilitate trades on behalf of brokerage users, or the persons providing or controlling such AI agents, be required to act in the best interest of the user, disclose conflicts of interest, maintain records of transactions and recommendations, safeguard customer information, or assume fiduciary or other legal duties with respect to user funds and investment decisions?” the lawmakers also asked.
Legislators asked Atkins to respond by July 31. The SEC did not respond to requests for comment prior to publication.
According to Joshua Broaded, the head of AI, Advisory Services, at the ACA Group, a compliance consulting firm, the focus on agentic tools stems from the advent of OpenClaw, a free and open-source AI agent that users can install on their computers.
If someone wanted that agent to operate on their brokerage platform, Broaded told Wealth Management that the platform likely never would have known. Some platforms (including the aforementioned Robinhood) are responding by creating frameworks to allow investors’ agents to interact with the platform.
As agentic tools continue to develop, Broaded said some brokerage platforms would remain more conservative in creating frameworks for agents to use.
However, other firms may see it as a “competitive advantage,” particularly in the context of a summer promising intial public offerings of massive AI-forward firms like Anthropic and OpenAI, which will leave thousands of AI-savvy shareholders with a sudden influx of investible assets.
“If you’re a brokerage firm and you’re signaling that you’re welcoming that, and you’re enabling trading strategies that leverage agents, it may be a way to pick up additional customers and enable something customers are interested in having,” he said.
Agentic tools (and the safeguards they employ) have increasingly been a focus for lawmakers in the wake of the unveiling of Anthropic’s Mythos model (as well as similar ones from competitors like OpenAI), which purportedly can rapidly probe (and find) cybersecurity vulnerabilities. Anthropic initially opted not to release Mythos to the public, citing that it was too powerful for general use. However, in late April, Anthropic investigated claims that a small group of individuals had accessed the secret Mythos model without the necessary permissions (though there was no evidence that malicious actors had gained access).
The breach underscored the greatest risk facing industries like financial services (and the registered brokerage platforms at the center of lawmakers’ concerns). According to Joshua Pantony, the CEO of Boosted.ai (an AI platform for large investors and advisors), third-party AI vendors may pose the largest risks for financial firms.
“To be clear, the Mythos breach didn’t happen because Anthropic got hacked,” Pantony said. “It happened through a third-party vendor on the day of launch. Every financial firm running AI right now should be doing one thing: mapping exactly what data their AI vendors have access to and what the damage looks like if that vendor gets hit. Most firms can’t answer that question cleanly, but that’s what we see as the actual exposure.”
For Broaded, the biggest compliance-related question remains who should be held responsible for the recommendation. Traditionally, a brokerage platform would argue that its human customers make trading decisions, and the platform merely takes and fulfills those orders. But the use of an agent muddies the clarity of that explanation.
“They’ve created the rails, they’ve invited the agents in, but it’s ultimately the customer that’s using them. So, I think it’s an open question,” he said. “It’s something that the SEC and perhaps lawmakers will need to decide; what sort of protections there should be for the invested public around AI-powered trading.”
When it comes to lawmakers’ concerns about AI vendors, Broaded said it was important to remember that there is a wealth of open-source options for savvy AI users, in addition to the largest frontier AI model providers such as Anthropic, OpenAI and Google DeepMind.
“It’s hard to predict the future,” he said. “But it would not surprise me if the AI labs come under some pressure to limit the ability of their models to make investment decisions, and that any controls that are built there are likely going to be able to circumvented by people who are technically and financially quite fluent.”
