Robinhood Lets Users Build Their Own AI Trading Agents, Powered by OpenAI or Anthropic

MENLO PARK, Calif. — Robinhood introduced Robinhood Agents on Sept. 29 at its HOOD Summit, a feature that lets customers build AI agents directly inside the app to research markets, develop trading strategies and execute trades on their behalf. Users can choose which model powers their agent, selecting from options offered by OpenAI or Anthropic rather than a single in-house model.

The company is rolling the feature out gradually to eligible U.S. customers rather than switching it on for its full base all at once. As part of the launch, Robinhood said OpenAI's GPT-Luna model would be available to users at no charge through the end of 2026, an introductory offer likely intended to seed adoption before any usage-based pricing takes effect.

Letting an AI agent execute trades autonomously is a meaningfully different proposition than the research-assistant chatbots that brokerages have already normalized. A tool that only summarizes a filing or explains an options strategy carries little downside if it gets something wrong; a tool empowered to place and manage trades on a customer's behalf raises the stakes of any model error, hallucinated data point, or misread instruction directly into real money moving in a real account.

Robinhood has not disclosed the specific guardrails placed on agent-initiated trades — order size limits, confirmation requirements, or the extent to which a human must approve an action before it executes — and this publication has not independently verified those controls. Customers weighing the feature should look for exactly that kind of detail in Robinhood's own documentation before granting an agent standing authority over an account, rather than assuming the same review process applies as when a person places an order manually.