Robinhood launches autonomous AI agents for trading and credit card purchases
Robinhood rolled out beta features allowing AI agents to autonomously execute stock trades and credit card purchases on behalf of customers through separate accounts and cards. FINRA identified AI agents as a new supervisory risk area, warning of potential for uncontrolled losses, data leaks, and decisions harmful to investors or markets without adequate safeguards.
What happened
Robinhood launched a beta feature allowing customers to connect AI agents like Anthropic's Claude to separate investment accounts via the Model Context Protocol, enabling agents to autonomously buy and sell stocks, read account positions and balances, and execute trades without explicit user approval for each transaction.
The company also introduced an Agentic Credit Card feature that lets AI agents make autonomous purchases using a virtual card with user-defined spending limits, such as booking flights or restaurant reservations.
Robinhood stated that users are responsible for all trades executed by agents and that the feature carries significant risk including possible total loss of investment.
Reported impact
FINRA classified AI agents as a new risk area, warning that agents could act without human approval, exceed user intent, make difficult-to-follow decisions, leak sensitive data, lack expertise for complex financial tasks, and be manipulated by poorly tuned reward functions toward decisions that harm investors or markets.
Index analysis
Robinhood implemented safeguards including separate sub-accounts limiting agent access to dedicated funds, real-time trade notifications allowing users to monitor activity, instant agent disconnection capability, preview functionality for certain orders, and fraud monitoring allowing support staff to trace executed instructions.
The feature is rolling out gradually in beta, currently supporting stock trading only with options, crypto, and other asset classes planned for later phases.
Governance controls
FINRA recommended that brokers build in safeguards, track agent actions, and determine where human oversight is needed.
Robinhood's risk disclosures warn that agentic trading is difficult to monitor or stop in real time and may not be appropriate for all investors.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.