Why Is the SEC So Silent About Agentic Trading?
Congress asked the SEC urgent questions about Agentic Trading, but the regulator’s silence leaves accountability, investor protection, and market risk unresolved.
On June 23, 2026, 8 House Democrats sent the Securities and Exchange Commission a letter about agentic AI trading. Led by Reps. Bill Foster of Illinois and Brad Sherman of California, the letter posed 13 questions to SEC Chairman Paul Atkins and asked for answers by July 31. The deadline passed 2 weeks ago. As of this writing on August 13, the Commission has issued no public response to the letter that I can find, and Atkins's public statements page lists nothing addressing the inquiry since the deadline. One clarification before going further: the SEC has not been silent about agentic finance. It has been silent about these 13 questions. The difference between those 2 silences is the story.
The silence is worth examining, because the letter is one of the clearest federal attempts yet to define the regulatory questions raised by agentic trading, the arrangement in which retail investors hand their brokerage accounts to autonomous AI agents. It is not the first: the SEC's own 2023 predictive data analytics proposal addressed conflicts in technology-driven investor interactions, and the letter pointedly asks the Commission to revisit it. Retail platforms now let customers connect AI agents that research, decide, and execute trades on their behalf. The technology moved from demo to deployment in under a year. The questions about who is responsible when it goes wrong have not moved at all.
What the 13 questions actually ask
Read as a list, the questions look procedural. Read as a group, they form an argument. I count 6 themes.
What did the SEC know, and what did it bless? Questions 1 and 2 ask whether the Commission consulted with brokerage platforms and AI firms before agentic trading launched, and whether it granted any formal approval, no-action relief, or interpretive guidance along the way. These are discovery questions. If the answer is "none of the above," then the most consequential change in retail market structure since commission-free trading arrived with no regulatory review at all.
What guardrails exist? Questions 3 and 5 ask what the SEC requires or even recommends: funding limits, position limits, order-size limits, approval procedures, transaction logs, restrictions on data access. Anyone who has built enterprise AI systems will recognize the list. It is the standard control inventory for any agent that can spend money, and the letter is asking whether the nation's market regulator has required any of it.
Who is responsible? Questions 4, 6, 7, and 8 form the legal core. Does an AI developer bear liability when its agent trades badly? Does a broker-dealer escape its obligations under Regulation Best Interest, supervision, and cybersecurity rules by classifying the agent as a "third-party tool" the customer chose to connect? When does an AI agent, or the company behind it, have to register as a broker, dealer, or investment adviser? Must the agent act in the user's best interest, disclose conflicts, keep records? The third-party-tool question is the sharpest of the 13. The lawmakers' stated worry is that platforms could use that label to sidestep the investor protections that would apply if a human employee did the same work.
Can the technology comply at all? Questions 9, 10, and 12 are the ones I find most interesting, because they are not really legal questions. They ask whether the SEC has analyzed whether generative models can meet suitability, disclosure, and fraud-prevention obligations in the first place, what rules govern the sale of customer data by AI developers, and whether the Commission has assessed the risk of agents acting on hallucinated information, outdated information, manipulated information, prompt-injection attacks, or social media rumors. That last list reads like the threat model section of an internal AI governance document, which is exactly what it is. Congress is asking the SEC whether it has one.
What happens when the agents agree with each other? Question 11 raises herding: similarly trained agents, prompted by similar users, converging on the same trades at the same time. Markets have seen correlated automation before. Portfolio insurance amplified the 1987 crash, and quant funds deleveraging together produced August 2007. Agentic herding would be the retail version, distributed across millions of accounts and coordinated by nothing more sinister than a shared base model.
Does the SEC even have the authority? Question 13 asks whether existing securities law suffices or whether Congress needs to act. It is the polite version of an uncomfortable question: is the referee silent because it chooses to be, or because the rulebook does not cover the sport?
Why silence is the expected outcome
Here is the deflationary reading, and it deserves a fair hearing. A congressional oversight letter is a request, not a subpoena. All 8 signatories are Democrats on the House Financial Services Committee, and none of them holds a gavel. Agencies routinely answer minority-party letters late, quietly, or never, and the reply, when it comes, is often delivered to the members without ever reaching the public docket. Two weeks past a self-imposed deadline is not a scandal by Washington standards. It is a Tuesday.
The SEC also has a preferred channel for answering questions like these, and it is not correspondence. Atkins did more than signal awareness. In May he devoted a section of a speech specifically to "AI & Agentic Finance," acknowledged opacity, rapidly propagating errors, and systemic vulnerabilities, and articulated a principle: regulated firms remain responsible for the outcomes of the tools they deploy. His July 7 statement on the 2026 regulatory agenda covers innovation, crypto assets, tokenized securities, and on-chain trading, though it announces no agentic trading rulemaking. When the Commission's answer to these 13 questions arrives, it will arrive on the Commission's schedule, and possibly never as a reply to this letter.
Yet the May principle does not answer the lawmakers' central question. "Firms are responsible for the tools they deploy" resolves the easy case, where the brokerage builds and ships the agent itself. The letter is about the hard case: the brokerage provides the account, a third-party developer supplies the intelligence, and the customer connects the two and grants the agent authority to act. Who deployed that tool? Question 6 lives exactly in that seam.
Why the silence still matters
The problem is what happens in the meantime. Rulemaking runs on a clock measured in years. Agentic trading is deploying on a clock measured in months. Readers of my governance writing know the lesson that keeps recurring, from baseball's wearable rules to federal data reporting: the leverage to set terms evaporates once the systems are running. Every month of regulatory silence is a month in which the default answers to all 13 questions get written by product teams, terms-of-service lawyers, and the accumulating weight of things already shipped. By the time a rule proposal arrives, "the agent is a third-party tool and we bear no responsibility for it" may not be a legal theory anymore. It may simply be how the industry works.
So the honest answer to my own title has 2 parts. The SEC is silent because nothing compels it to speak, and because agencies prefer to answer hard questions through rulemaking they control. But an unanswered letter is still an answer of a kind. The problem is not that the SEC has never thought about AI; Atkins has thought about it out loud, at length. The problem is that his stated principle, firms are responsible for the tools they deploy, may break down exactly where the agentic architecture divides deployment, control, advice, and execution among different parties. Eight members of Congress asked where responsibility falls when the broker provides the account, a third party supplies the intelligence, and the customer gives the agent authority to act. Nobody has yet said. The 13 questions are good ones. Somebody should answer them before the market does it by default.
Further Reading
- Foster, Sherman Seek Regulatory Clarity on Agentic AI Trading - the press release announcing the letter.
- The letter itself (PDF) - all 13 questions in full.
- Is agentic trading safe? What the SEC inquiry means for investors - Stacker's explainer on the inquiry.
- Remarks at the SCSP AI Expo - Chairman Atkins's May 8, 2026 speech, including the "AI & Agentic Finance" section.
- Predictive Data Analytics proposal (S7-12-23) - the 2023 rulemaking the letter asks the Commission to revisit.
- SEC chair Atkins signals rule changes for onchain markets and AI-driven finance - CoinDesk on the Commission's rulemaking posture.
- Statement on the 2026 Regulatory Agenda - Chairman Atkins, July 7, 2026.