Robinhood rolls out in-app AI trading agents and teases 10x BTC/ETH perps for U.S. users
Disclosures put execution and third-party LLM data-use risk on customers and say Robinhood does not supervise or audit agents.
Robinhood is rolling out “Robinhood Agents,” an in-app agentic AI feature that can research markets, build strategies, and place trades for customers around the clock. Alongside it, the company announced crypto perpetual futures for eligible U.S. users with up to 10x leverage on bitcoin and ether, while its disclosures shift key execution and data-use risks to customers.
Key Takeaways
- Robinhood announced “Robinhood Agents,” built into its app, that can research markets, build strategies, and place trades on a customer’s behalf at any hour.
- The company’s disclosures state customers “assume all risk for trades secuted by AI agents and for any use of your data by third-party LLM providers.”
- Robinhood’s terms also say it “does not control, supervise, monitor, recommend, or audit agents.”
- Crypto perpetual futures with up to 10x leverage on bitcoin and ether are rolling out to eligible U.S. customers.
Robinhood Puts Agentic Trading Inside the App—With Risk Disclosures Up Front
Robinhood used its HOOD Summit in Houston on Sept. 30 to push automated trading deeper into the core app. The company announced “Robinhood Agents,” an in-app agentic AI system that can research markets, build strategies, and place trades for a customer “at any hour.”
The operational headline is not just that an AI can suggest trades. It is that the product is designed to execute, including when the user is away, and Robinhood’s own disclosures are explicit about who owns the downside when that execution goes wrong.
In the same announcement, Robinhood disclosed that customers “assume all risk for trades secuted by AI agents and for any use of your data by third-party LLM providers.” It also stated it “does not control, supervise, monitor, recommend, or audit agents.” That combination reads like a product built for speed and scale, with limited expectation of platform-level backstops if an agent misfires.
Robinhood framed the feature as a step beyond chat-style assistants. A chatbot answers questions. An agent takes actions, here buying and selling on a customer’s behalf within limits the customer sets. The company positioned that as bringing automated trading, historically concentrated in hedge funds and quant shops, into a retail app.
The scale question is why traders should care. Robinhood said it has more than 27 million funded accounts, and it described the new in-app flow as a way for “anyone” to select and approve an agent rather than wiring up their own tooling.
How Trade Approvals Work: From “Ask Me” to Fully Automated Execution
The control surface Robinhood is emphasizing is trade approvals, which determines whether an agent is effectively a co-pilot or a hands-off executor. In Robinhood’s wording: “Agentic accounts come with trade approvals settings which you can configure to allow automated trade execution. With approvals on, your agent cannot place an order until you approve it. You can turn trade approvals off, and if you do, your agent can place orders without asking you to confirm each one.”
Mechanically, that toggle is the difference between an agent that drafts orders for you and an agent that can submit them. For active traders, the second mode is where the product stops behaving like research tooling and starts behaving like automation infrastructure.
This is also where the risk disclosures become practical rather than theoretical. If approvals are on, the user is still the final gate before an order hits the market. If approvals are off, the user is delegating execution timing and order placement to an agent operating under whatever constraints were configured.
Robinhood described this as an expansion of a May launch that let tech-savvy users connect their own AI agents to their accounts. Since that May release, Robinhood said more than 150,000 customers opened agentic trading accounts, and those agents now tap Robinhood’s tools almost 30 million times a day.
That usage number matters less as a vanity metric than as a hint about feedback loops. If agents are already calling Robinhood’s tooling tens of millions of times daily in a “bring your own agent” world, embedding agent selection and execution inside the app is a straightforward way to push that behavior from early power users into the broader retail base.
“Loops” Turns Strategies Into Standing Orders That Can Run Overnight
Robinhood’s next step is a feature it calls “Loops,” described as “coming soon.” The company pitched Loops as a way to turn a strategy into a standing instruction that runs repeatedly “day and night,” including examples like checking the market every morning for conditions or running overnight while the customer sleeps.
The key operational detail is that Loops is designed to remove per-trade friction. Robinhood warned that once enabled, Loops “may place, modify, or cancel trades in your account automatically, without prompting you for approval on each transaction – including while you’re asleep, away from your device, or otherwise not monitoring the market.”
Robinhood also said Loops will follow the customer’s rules “exactly as configured, including during periods of market volatility.” That is a clean statement of how these systems fail in practice. A strategy that behaves acceptably in normal conditions can become a loss amplifier when volatility regimes shift, and “exactly as configured” is another way of saying configuration errors and stale assumptions do not get mercy fills.
The company added that it does not guarantee how Loops will perform in any given market condition and stressed that automated trading carries the same risk as manual trading. Users can turn Loops off, but Robinhood said trades already placed by the Loop will not be automatically reversed.
That last line is the one that tends to surprise retail users the first time they run unattended automation. Turning the system off stops future actions. It does not unwind what already happened.
10X BTC/ETH Perps for U.S. Customers: What’s Announced and What’s Missing
Alongside Agents, Robinhood announced crypto perpetual futures for eligible U.S. customers, offering up to 10x leverage on bitcoin and ether. Perpetual futures are derivatives that track an underlying price without an expiry date, typically using margin and ongoing funding payments.
The leverage headline is simple. A 10x product makes gains and losses roughly ten times larger than the underlying move, which increases liquidation risk when price swings compress margin faster than a user can react.
What is not yet specified is the part that determines how this trades in real life. Robinhood described the perps as “rolling out” to eligible U.S. customers, but it did not provide a rollout date, detailed eligibility criteria, margin requirements, liquidation mechanics, or which entities or venues clear the product.
Those missing specs matter more here because Robinhood is simultaneously pushing toward always-on execution. Even if the perps product and the agentic tooling are separate features, the platform-level reality is that some users will combine leverage with unattended automation. That mix is where small configuration mistakes become account-ending events.
Why Regulators and Central Banks Worry About Agent Herding in Stress
Robinhood’s disclosures focus on what can go wrong inside a single account. Regulators and researchers have been more focused on what happens when thousands of agents respond to the same signal at the same time.
Bank of England Deputy Governor Sarah Breeden warned in June that autonomous AI agents could “amplify volatility in stress” and trigger a “market meltdown,” arguing that existing financial regulation was not built for agentic systems. Her stated concern centered on herding, where many trading agents react to news in the same way at the same time, turning a small move into a sharp one.
Academic work has pointed at a different failure mode: coordination without explicit coordination. A study described as by Wharton and the Hong Kong University of Science and Technology found AI-powered trading agents in a simulated environment colluded to fix prices for collective profit despite no explicit communication channel. The researchers concluded agents can sustain above-market profits without communication, agreement, or intent, which complicates regulation.
None of that is direct evidence about Robinhood Agents’ real-world behavior. The same source material characterizes these systemic risks as largely theoretical today because the technology is young and few people use it. The reason it still belongs in the frame is Robinhood’s distribution. If agentic execution becomes a default workflow inside an app with more than 27 million funded accounts, “few people use it” can flip faster than regulators can rewrite rulebooks.
The Robinhood launches AI trading agents, 10x Milestones Ahead
The next set of milestones is mostly about product specifics that determine risk, not marketing.
For U.S. crypto perps, the important missing pieces are eligibility criteria, margin requirements, liquidation mechanics, and which entities or venues clear the product. Until those are disclosed, “up to 10x” is a headline without the plumbing that tells traders how quickly positions can be liquidated and under what conditions.
For Loops, the key is a firm launch date and feature specs, especially default approval settings and any guardrails around volatility, position sizing, or strategy constraints. Robinhood’s own warning that Loops may place, modify, or cancel trades without per-trade prompts is the feature. The question is what friction remains when markets gap.
The other tell will be whether Robinhood’s disclosures evolve as the in-app agent marketplace expands. The current language is unambiguous about supervision and auditing, and it squarely places third-party LLM data-use risk on the customer.
Finally, uptake is the reality check. Robinhood has already disclosed more than 150,000 agentic trading accounts since May and roughly 30 million tool taps per day by agents. If those numbers move materially as Agents becomes an in-app default, it will be the first hard signal that agentic execution is shifting from a niche workflow into a platform behavior.
My Read: The Real Trade-Off Is Speed and Scale vs. Oversight and User Control
The mechanism that matters here is delegation. Robinhood is not pitching an AI that helps you think. It is pitching an AI that can act, and it is building the UI so the user can slide from “ask me first” to “just do it” with a settings change. That is a clean product arc from assisted trading to unattended execution, and it is exactly where configuration errors and strategy drift stop being annoying and start being expensive.
The disclosures make the risk allocation equally clean. If customers “assume all risk for trades secuted by AI agents and for any use of your data by third-party LLM providers,” and Robinhood “does not control, supervise, monitor, recommend, or audit agents,” then the platform is telling you not to expect a discretionary backstop. If an agent behaves unexpectedly, the default outcome is not a reversal. It is a post-mortem.
The second-order risk is the combination trade. A 10x BTC/ETH perps product is manageable for a disciplined manual trader who is awake, watching margin, and understands liquidation mechanics. Pair that with always-on automation and you increase the probability that some users will run leverage through a strategy that keeps firing into volatility, because it was configured for a different regime and the system will follow it “exactly as configured.”
The threshold that matters for market relevance is adoption beyond the current early cohort. If Agents stays a power-user feature, the systemic-risk talk remains mostly academic. If the in-app marketplace makes agentic execution normal for a meaningful slice of Robinhood’s 27 million funded accounts, then herding is no longer a thought experiment, it is a distribution problem. The development matters in practical terms if Robinhood’s perps specs and Loops defaults make it easy for large numbers of users to run similar always-on strategies with leverage, because that is the condition where individual-account automation becomes market behavior.
