Skip to content
Agentic AI

AI Agents Trading Accounts Without Approval: Interactive Brokers Goes Autonomous, Robinhood Passes 150,000 And The FCA Names The Risk

On 21 October 2026 Interactive Brokers launched IBKR Autonomous Trading, letting connected AI platforms place orders across more than 75,000 instruments without the client approving each trade. A Jefferies report summarised on 27 October counts more than 150,000 agentic accounts at Robinhood. A day after the launch, the FCA's chief executive warned about agents coordinating into market manipulation. What changed, and what brokers now have to build.

AlchmAI Editorial7 min read

75,000+

Listed stocks, options and ETFs an IBKR AI partition can trade without per-trade approval (The TRADE, 22 October 2026)

150,000+

Robinhood customers with agentic trading accounts as of 29 September 2026 (Jefferies, via Investing.com)

~30m

Times a day agents use Robinhood's tools, per the same Jefferies report (Investing.com, 27 October 2026)

4

AI platforms IBKR already connects to through MCP: ChatGPT, Claude, Grok and Perplexity (The TRADE)

On 21 October 2026 Interactive Brokers launched IBKR Autonomous Trading, which lets eligible clients authorise a connected AI platform to find trades and place orders without approving each one. The AI trades from a dedicated partition funded with cash moved from the client's main account, across more than 75,000 listed stocks, options and ETFs. For brokers, AI agents trading accounts have moved from a supervised experiment to a product that runs while the client is asleep.

The launch did not arrive alone. A Jefferies report summarised by Investing.com on 27 October counts more than 150,000 agentic trading accounts at Robinhood and new agent features at E*Trade, Schwab, Coinbase and Tastytrade. A day after the IBKR launch, the FCA's chief executive Nikhil Rathi told the City Dinner at Mansion House that one judgement regulators now face is "when agentic coordination becomes market manipulation".

We covered the first wave in August, when brokers opened MCP servers and kept a human approval step. This is the material change: at the largest broker to do it so far, the approval step is now optional.

What Did Interactive Brokers Launch On 21 October 2026?

According to The TRADE's report of 22 October 2026, IBKR Autonomous Trading lets clients describe a strategy in natural language. The AI then watches the market and places supported orders that meet those criteria. Chief executive Milan Galik said clients define their strategies and safeguards and that the AI executes trades "within those parameters". He framed the launch as continuity: "For decades, Interactive Brokers has used automation to give investors greater access, efficiency and control."

The new feature extends IBKR's existing AI integration, which connects client accounts through MCP to ChatGPT, Claude, Grok and Perplexity. Jefferies notes that IBKR widened that integration in July from three models to any language model built on the MCP standard. The TRADE's report does not specify order limits, cash caps or other risk controls inside the partition, and IBKR's public description leaves the safeguards to the client.

How Far Have AI Agents Trading Accounts Spread Across Brokers?

Further than most broker product roadmaps assumed a year ago. The Jefferies report, as summarised by Investing.com on 27 October 2026, lists the moves since its first look in June. E*Trade launched MCP connectors on 8 October that link live accounts to third-party AI models. Schwab began rolling out a conversational agent in October and is testing MCP-based trading tools with one model provider. Coinbase added US stocks and ETFs in September and index and crypto perpetuals in October. Tastytrade plans fully autonomous agents that execute multi-asset strategies.

Robinhood is the scale case. More than 150,000 customers had opened agentic trading accounts by 29 September, and their agents call Robinhood's tools nearly 30 million times a day. At the end of September Robinhood unveiled Robinhood Agents, bots that analyse markets, build strategies and execute trades inside the app, and previewed Loops, which would let agents carry out standing instructions around the clock. Its own line was: "The stock market never sleeps, and neither should your portfolio." Chief executive Vlad Tenev described the tools as ones "once reserved for hedge funds, big banks, and quant firms".

Jefferies' headline says these tools are adding to market volumes. The report, as reported, gives no volume figures, so we treat that as the analysts' reading of direction rather than a measured result.

What Is The Risk When Agents Trade Without Approval?

Two risks, one for the client and one for the market. For the client, the approval step was the last place a person could catch a misread instruction, a stale price or a model that drifted from the strategy it was given. Without it, the controls have to sit in the account: position limits, loss limits, instrument lists and a kill switch that work whether or not the model behaves.

For the market, the risk is many agents reading the same signal at the same moment. Rathi's speech on 22 October 2026 named it directly. He said the opportunities are vast, "But the threats to market integrity are as serious as I have seen them", and listed "Unprecedented volatility, concentration risks, new forms of market abuse and rampant cross-border financial crime". He also said personal AI agents are expected to reach UK consumers soon. A broker that offers autonomous trading to UK clients is now offering a product the regulator has already flagged in a set-piece speech.

None of the US launches is available to UK retail clients in the same form today, and the FCA has not published rules specific to agent trading. Its stated approach is to rely on Consumer Duty and the Senior Managers regime. That makes the design choices, not a rulebook, the place where a firm shows it took the risk seriously.

Our View: Approval Was The Control. Now The Controls Must Live In The Account

The uncomfortable point for brokers is that removing per-trade approval does not remove responsibility; it moves it. When a client approved every order, the client's click was the audit trail. When an agent trades on its own, the broker's platform is the only place that can prove which instruction the agent was following, which limits applied, what the agent saw and why it stopped. A partition with cash in it is a ring fence for money. It is not yet a ring fence for behaviour.

The firms that win this market will be the ones whose agent accounts are boring to supervise: limits enforced by the platform rather than the prompt, every agent order tagged as agent-originated, and a record a compliance lead can read the morning after a volatile session. The firms that treat autonomy as a feature toggle will learn about agent herding from their own client complaints.

“Taking the human approval out of an AI agent's trades does not take the responsibility out. It moves it into the broker's platform, and the platform has to be able to prove it.”

What It Means For Trading And Fintech Firms

  • Brokers and CFD providers: if you plan agent access, enforce limits outside the model: maximum position, daily loss, permitted instruments and an off switch the client and your dealing desk can both use. Tag every agent-originated order so surveillance and complaints teams can tell it apart from a manual one.
  • Prop trading firms: decide now whether evaluation accounts may be driven by agents. If they may, your rule engine needs to treat agent orders the same way it treats human ones, with the rule version and the agent's instruction stored next to every breach.
  • Compliance leads: map an autonomous agent product against Consumer Duty before launch. The question is not whether the agent is clever but whether the client understands what it may do with their money while they are not watching.
  • Fintech and trading startups: build on the open standard. MCP is the common interface across IBKR, Robinhood and E*Trade, so a well-scoped MCP server with read-only tools by default is the cheapest way to be agent-ready without being agent-exposed.
  • Research and quant teams: start measuring agent flow. If agent orders are tagged, you can see whether they cluster around the same signals, which is exactly the herding risk the FCA described.

The Bottom Line

AI agents trading accounts crossed a line in October 2026: Interactive Brokers now lets connected AI platforms trade more than 75,000 instruments without per-trade approval, Robinhood counts more than 150,000 agentic accounts making about 30 million tool calls a day, and the FCA's chief executive has publicly asked when agent coordination becomes manipulation. The technology is ready; the supervision is not yet standard. AlchmAI, a London trading technology and AI studio, builds agent access and the limits, logging and kill switches around it for brokers and prop firms, and the lesson of this month is that those controls are now the product, not an add-on.

References & Further Reading

Agentic AIAI agents trading accountstrading automationInteractive Brokersautonomous tradingModel Context ProtocolAgentic AI London
Share Email
AI

AlchmAI Editorial

Research and analysis, London

The AlchmAI team writes about the markets, technology and regulation we work with every day. We build trading platforms, real-time charts and AI analysis tools for brokers, prop firms and fintech teams from our office in Mayfair, London. Every article lists its sources. Nothing we publish is investment advice.

This article is general information and commentary. It is not investment advice or a recommendation to buy or sell any investment. Important information

Keep reading

All articles