Picture a litigation partner who already uses an AI assistant to summarize depositions and draft first versions of motions. One evening he notices that his brokerage app now offers to connect that same kind of assistant to his account. It could read his positions, explain them, and place trades. At one firm, if he tells it to, it could place those trades without asking him first.

A year ago, asking an AI chatbot about your portfolio meant copying numbers into a chat window and hoping it understood them. In 2026, three brokerage firms began letting AI connect directly to customer accounts. I trade every week, and I’m interested in this technology and wary of it in equal measure. Here is what each firm allows, where the real risks sit, and the ground rules I’d set before giving any software access to money.

What “AI agent” means here

An AI assistant is a program you converse with in plain English, such as Claude or ChatGPT. It becomes an agent when it is allowed to take actions on your behalf instead of only answering questions: looking up your balances, filling in an order ticket, or sending that order to the market. The question that matters for an investor is how far along that chain the software is allowed to go before a human says yes.

Three firms, three levels of control

Interactive Brokers: you approve every order

On June 1, 2026, Interactive Brokers let clients link their accounts to Claude. The assistant drafts orders for stocks and exchange-traded funds, and the client approves each one before it goes to market. Nothing executes on its own. This is the most cautious of the three setups, and it suits someone who wants help with research and paperwork while keeping every decision, much as a senior partner reviews a junior associate’s draft before it is filed.

Robinhood: an agent that can act alone

Robinhood announced Agentic Trading on May 27, 2026, and it is still in an early testing phase, known as beta. You connect an outside AI agent through a technical connection Robinhood calls an MCP server, a standard way for AI tools to plug into other services. Supported tools include Claude, ChatGPT, Codex, Cursor, Grok, Perplexity and others.

The design includes some sensible protections:

  • The agent can trade only in its own dedicated account, separate from your main one. You need a primary account in good standing, and Robinhood allows up to 10 individual accounts per customer.
  • The agent can answer questions about your portfolio and place orders. Its tools cover stocks, options and crypto, though Axios reported at launch that trading was limited to stocks, with more to come.
  • You get an alert every time the agent trades, and you can pause or switch it off instantly.
  • You can review what the agent plans to do. If you tell it not to ask, it can trade without confirming with you.

Two details deserve close attention. First, although the agent can trade only in its own account, it can see all your Robinhood accounts: positions, balances and history. Second, Robinhood is clear that you are responsible for the agent’s trades. The company warns that you could lose your entire investment and that agents can misread instructions or act on outdated information.

Robinhood paired the launch with an AI-enabled credit card feature tied to its Gold Card. You set a spending limit and can require your own approval for each purchase.

Public: agents for event contracts

On September 23, 2026, Public launched AI agents that can trade event contracts listed on Kalshi. An event contract pays a fixed amount if a specific event happens, such as a particular inflation figure or an interest rate decision, and nothing if it doesn’t; these contracts are regulated by the Commodity Futures Trading Commission. It is a narrower use than Robinhood’s, aimed at a market where prices react quickly to headlines, which is exactly where software that reads the news fast has obvious appeal.

The do-it-yourself route

Beyond the firms’ own offerings, freely available open-source programs can research ideas, test trading strategies against past market data and connect to brokerage accounts. Some are impressive. All of them deserve the same caution as any software written by a stranger.

My rule is to treat an AI-generated strategy the way an engineer treats an untested design. A test against historical prices, called a backtest, can look brilliant and still rest on a flaw nobody has spotted: a programming error, an unrealistic assumption about the prices you’d actually get, or information that accidentally leaked from the future into the test. The AI that wrote the strategy is a poor judge of whether it works.

Where the risks actually sit

Most of the likely problems are ordinary ones:

  • Misread instructions. “Trim my tech exposure” can mean very different things. An agent may choose an interpretation you never intended.
  • Stale information. An agent acting on a price or headline that is an hour old can make a sound decision on bad facts.
  • Speed. A person makes a mistake once. An automated process can repeat it dozens of times before anyone notices.
  • Borrowed money. In an account that borrows against your holdings, or in futures, an error is magnified. Futures losses can exceed the money you put up.
  • Privacy. Giving an agent read access to every account means a great deal of financial detail flows to the AI provider. Know what you are sharing.

None of this is unique to AI. Bad instructions and automated mistakes have cost traders money since the first computerized trading programs. What’s new is how easy it has become to hand the controls to software.

Three questions for any firm

Before connecting an agent anywhere, find the answers to three questions in the firm’s terms: Which accounts can the agent see? Which accounts can it trade? What does it take to switch off its confirmations? At Robinhood, the answers are all of your accounts, one dedicated account, and a single instruction telling it not to ask. At Interactive Brokers, every order waits for your approval.

Ground rules before you connect anything

If you want to experiment, these are the guardrails I’d use:

  • Use a separate, small account. Fund it with an amount you could lose without it mattering. Robinhood’s dedicated agent account is built for this.
  • Practice first. Most platforms offer a simulated account with pretend money, known as paper trading. Run the agent’s ideas there for a few weeks before using real money.
  • No borrowed money. Keep the agent’s account a cash account, one that can’t borrow against its holdings, so its mistakes can’t grow past what’s in it. The distinction matters more than ever under the new day-trading rules, which I cover in Cash Account or Margin Account? What the New Day-Trading Rules Change.
  • Set a loss limit. Decide in advance how much of a decline will make you switch the agent off, and keep to it.
  • Keep your main accounts unconnected. Long-term savings and retirement money stay out of reach.
  • Review every trade at the start. Keep confirmations on until you understand how the agent behaves. Approval-only setups like Interactive Brokers’ make this the default.

My view

The draft-and-approve model can save real time. An assistant reads your positions, explains them in plain English and writes up the order, and you still approve every decision. Fully autonomous trading is a different proposition. It can work for a disciplined person running a narrow, well-tested strategy in a small account. For long-term wealth, I’d keep a human hand on every order.

AI agents are one piece of a wider set of changes at brokerage firms this year, which I lay out in The Great Broker Shake-Up of 2026. I track each launch by date on BrokerNimble’s what’s-new page, and my full guide to choosing and using a broker covers how to weigh features like these against the costs that matter every day.

Disclosure: Peter Agro runs BrokerNimble.com, an independent broker comparison site that plans to join broker affiliate programs and may earn referral fees in the future. Figures were checked in October 2026 and change often; confirm current terms with the provider. The attorney above is an illustrative example. This article is for information only and is not investment, tax or legal advice. Futures, margin and options trading involve substantial risk of loss, including losses greater than your deposit.