Robinhood has opened real brokerage accounts to outside AI agents, and the important part isn't the AI. It's who carries the loss when the agent gets a trade wrong.
Robinhood launched Agentic Trading on May 27, and the product does exactly what the name suggests. You can connect an outside AI agent, including Claude, ChatGPT, Codex, Cursor or Grok, to a dedicated Robinhood account and let it place stock orders with money you've moved into that account. No fantasy portfolio. No paper-trading sandbox. Real orders, real fills. Real loss of principal, too.
According to Robinhood's own support page, the agent can analyze your portfolio, look at buying power, assess concentration risk, build a strategy around a theme such as AI or semiconductors, and place orders through Robinhood's Trading MCP server. MCP is the Model Context Protocol, the open standard Anthropic pushed into the market so AI tools can talk to outside services and take actions. Robinhood didn't need to build a whole assistant from scratch. It built the socket.
That distinction matters. Robinhood's disclosure says agentic trading can execute trades without your direct input on each transaction, and that AI-driven strategies may perform poorly, misread instructions, act on stale information or behave unexpectedly. You are responsible for reviewing activity and monitoring positions. The company says it isn't responsible for losses resulting from agent-generated decisions.
That's the whole trade-off.
Robinhood has added guardrails, but you should read them as limits on damage, not protection from bad judgment. The agentic account is separate from your main portfolio, so the agent can only trade with the funds you put there. Robinhood says you get notifications when the agent trades, a real-time activity feed, profit and loss tracking, and a disconnect button. Those are useful controls. They don't change the basic deal: once you give the agent authority, the agent can act.
The credit card side follows the same pattern. Robinhood also announced an Agentic Credit Card, a dedicated virtual Robinhood Gold Card that an AI agent can use for purchases inside spending limits you set. Stock trading is the only live trading asset class in beta, while Robinhood says options, crypto, event contracts and futures are planned.
The Race Is Already Crowded
Robinhood says it has more than 27 million funded customers. That is why this launch lands differently from a developer demo. A feature that starts inside a mainstream brokerage can move from technical curiosity to consumer habit much faster than most finance products.
Coinbase moved quickly too. The company launched Coinbase for Agents on June 11, giving AI agents access to Coinbase accounts so they can trade and pay, operating within user-set limits. On June 16, Coinbase launched Coinbase Advisor, an AI-powered advisor available through Coinbase One. Coinbase's own terms say Coinbase Advisors, LLC is registered with the SEC as an investment adviser and with the CFTC as a commodity trading adviser, while also making clear the product is non-discretionary and every action requires approval.
That difference is important. Coinbase Advisor gives recommendations and can help execute after approval. Robinhood Agentic Trading is built around the possibility that an outside agent places orders without asking you each time, if you've configured it that way. One model keeps the human at the button. The other lets the machine press it.
eToro has been working the same idea from another angle. It announced Agent Portfolios in March, letting users connect agents to dedicated sub-portfolios with scoped API keys and a minimum starting allocation of $200. That's not a high bar. In July, eToro said its redesigned app would include agent-powered portfolios that trade inside their own sub-accounts. Interactive Brokers has also expanded AI integrations, adding ChatGPT and Grok in June for research, analysis and order-instruction workflows across equities, ETFs, options and futures.
Charles Schwab looks more cautious. Its May announcement covered AI-generated portfolio and market summaries for retail clients, designed to provide information rather than investment advice. Fidelity's public trading pages still describe tools for research, order entry and portfolio management, not a Robinhood-style outside agent placing trades on its own.
The Liability Question Is the Product
Vlad Tenev is not hiding the ambition. CNBC reported on July 2 that Robinhood's CEO said AI agents would soon match the capabilities of human traders, and that the goal is to give ordinary investors the kind of computation and speed that institutional and high-frequency trading firms have used for decades. Tenev has standing here. Before Robinhood, he worked on programmatic trading systems.
Frankly, that history cuts both ways. Automation did change Wall Street, but it didn't remove risk. It made risk faster and cheaper - and easier to scale. Put that in a retail brokerage app and the obvious question is no longer whether an AI agent can trade. It can. The question is whether most customers understand what they have authorized before the first bad order lands.
Robinhood's answer is to wall off the account and disclose the risks - then put the monitoring burden on you. Coinbase's answer, at least with Advisor, is to route advice through a registered adviser and require approval before action. eToro's answer is a dedicated agent portfolio. These are different legal and product choices, not just different interfaces.
For now, regulators don't need a theory paper to see the issue. They have brokerage accounts connected to outside AI systems, customer money moving through automated instructions, and disclosures that push responsibility back to the person who clicked connect. The agents are trading. The notifications are coming after the decision. The money is real.
Also read: Cursor Doubles Usage Limits Across Every Paid Plan as Rivals Close In • SkyPilot Raises $20 Million to Let AI Teams Shop Compute Across Every Cloud • Alibaba Says Qwen3.8-Max Beats Fable 5 on Price, but Skips the Benchmark