Robinhood rolls out OpenAI and Anthropic trading agents to roughly 29M customers
The in-app agents can run automated “Loops,” with guardrails and token-rate pricing that leaves cost and liability open.
Robinhood began rolling out in-app trading agents powered by OpenAI and Anthropic to all of its roughly 29 million customers starting the week of Sep. 29, 2026. The agents take plain-English instructions to research, build strategies, and execute trades, pushing agentic execution into a retail user base large enough to matter for flow.
Robinhood Brings AI Trading Agents to 29M Accounts
Robinhood is distributing AI trading agents to its full retail footprint. Starting the week of Sep. 29, the company said all of its roughly 29 million customers will have access to in-app agents powered by OpenAI and Anthropic.
Mechanically, this is a shift from “AI features” to delegated execution. Robinhood is positioning the product as a way for customers to instruct an agent in plain English to place trades, do research, and assemble multi-step strategies, with the rollout timed to its annual HOOD summit.
The launch also closes a loop Robinhood opened earlier this year. In May 2026, the company released an MCP tool, a connector framework that let technical users hook up their own agents to Robinhood’s trading platform. This week’s release targets nontechnical users directly inside the app, which is the part that can change retail behavior at scale.
Robinhood CEO Vlad Tenev framed the push explicitly around active trading. “Ownership doesn’t work without markets, and markets don’t work without traders,” Tenev said. “We’re making Robinhood the best place in the world for active traders by delivering tools once reserved for hedge funds, big banks, and quant firms.”
How “Loops,” Model Choice, and Guardrails Change Execution
In a product demonstration, users could name an agent and select which model runs it. The models shown were OpenAI’s GPT-6 Luna, OpenAI’s GPT-6 Sol, and Anthropic’s Opus 4.8.
From there, the interface is natural-language execution. Robinhood’s example of a basic instruction was “Buy $200 of Ford stock.” The more market-relevant feature is automation: Robinhood’s “Loops,” which it described as strategies that repeatedly check conditions on a schedule and trade when rules are met. “You can set a Loop to check the market every morning and execute a trade when certain conditions are met, or run a continuous overnight strategy to look for opportunities while you sleep,” Robinhood said.
Robinhood says it built guardrails to keep agents from behaving in unexpected ways, and the controls are mostly about isolating blast radius. The company described (1) a dedicated trading account for the agent, (2) user-set limits on how much the agent can trade at a time, and (3) an optional confirmation step that requires final approval before execution.
The other constraint is cost, and Robinhood is choosing a “try it free, then meter it” path. For the rollout period, Robinhood planned to offer the lower-end GPT-6 Luna for free until the end of 2026, while charging the standard token rate for OpenAI and Anthropic agents. Token-rate pricing means heavy iteration and research loops can become the real bill, even if single trades feel cheap.
Adoption Signals, Cost Friction, and the Liability Gray Zone
Robinhood is already pointing to early usage, but the metrics are still hard to map to risk-taking. The company said over 150,000 customers opened agentic accounts using the earlier technical tool introduced in spring 2026. It also said that as of late September 2026, various agents were transacting on its platform nearly 30 million times per day.
That “nearly 30 million times per day” figure is the one traders will want decomposed. Robinhood did not specify whether it refers to completed trades, API calls, research actions, or some mix, and the market impact depends on which bucket dominates.
The most obvious market-structure risk is strategy crowding. Robinhood itself raised the possibility of agents converging on similar signals, including a hypothetical scenario where agents “confer with one another” and move en masse into or out of an asset, which could amplify volatility or panic if malicious actors are involved.
Liability is the other unresolved edge. Robinhood executives argued hosting agents does not constitute providing financial advice, likening it to customers asking the internet or a friend, while also acknowledging the legal landscape is still evolving.
Near term, the cleanest adoption signal will be whether Robinhood discloses how many of its roughly 29 million customers activate agentic accounts, and how many run automated Loops versus one-off commands. The second is whether Robinhood updates the “nearly 30 million times per day” activity metric with a clearer definition of what counts as a transaction. The third is pricing behavior after the end-of-2026 free period for GPT-6 Luna, since token-rate billing can quietly cap research-heavy strategies even if execution remains cheap.
My Read: Agentic Retail Flow Is Now a Market-Structure Variable
The key market signal here is distribution, not capability. Retail has been able to automate trades for years, but putting agentic execution behind a native UI for roughly 29 million accounts raises the odds that similar playbooks get crowded quickly, especially once “Loops” become copyable patterns rather than bespoke scripts.
The threshold that matters is whether Robinhood can show that Loops are being run at meaningful scale, not just tested. If the company starts breaking out activation rates, Loop usage, and what “nearly 30 million times per day” actually measures, agent-driven retail flow stops being a narrative and becomes a measurable input into liquidity and volatility.
