Brokers open trading platforms to AI agents, shifting control

Interactive Brokers introduced autonomous trading in October. The feature gives eligible clients control over the limits, after which connected AI can place supported orders for them. Scalable Capital has also opened its platform to major AI assistants. Both moves bring AI closer to the brokerage account and the order path, rather than keeping it as a research feature outside the trading workflow.

That raises an infrastructure question for brokers: who controls the customer interface, and how much of the service can be delivered through another firm's agent? A broker may still hold the account and execute trades, but an outside assistant could become the place where customers compare products, examine portfolios and decide what to do.

AI tools can also simplify platform tasks such as screening instruments, analysing portfolios, interpreting earnings and automating routine execution. Brokers may apply similar systems to support, onboarding and everyday account queries. Those uses could reduce service workload, but the net benefit depends on integration, model, monitoring and compliance costs as well as any savings.

Where AI changes broker operations

  1. Service and customer retention. Automated support could help firms serving customers across languages and time zones, while assistants may appeal to technically sophisticated traders. The operational test is whether service costs fall without adding more expense in oversight and compliance, and whether customers stay longer as a result.
  2. Account and order integration. The stronger proposition is an assistant connected to account information, market data, execution and risk controls, rather than a chatbot added to an app. That integration raises questions about permissions, supported orders and the boundaries set for each client. Interactive Brokers' announcement provides one example of connected AI submitting orders under client-defined limits.
  3. Different models, different exposure. CFD/forex brokers could use AI for risk analytics, retention, trading tools and multilingual service, but more activity does not necessarily produce durable customer value and may increase conduct and compliance exposure. ETF brokers have to weigh easier research and automated investing against features that become less distinctive as price competition grows. Multi-asset firms may gain from cross-asset discovery and joined-up analytics, while taking on added complexity and cost.
  4. Value across the chain. Brokers can earn from spreads or commissions, assets held on-platform and reduced service costs. Model providers and agent developers may make money from API use, subscriptions, enterprise contracts or control of the customer interface. Investors could automate routine portfolio work and examine more instruments, but charges, transaction costs, poor decisions or excessive trading can outweigh any gains. Better analysis alone does not ensure better returns.
  5. Evidence, not adoption counts. Internal performance data is more useful for assessing whether AI pays than adoption surveys. A comparison also needs to account for the possibility that people who select AI tools were already more active or technically confident. Without that control, a broker cannot confidently attribute stronger retention or profitability to the tools themselves.

Regulators question who captures value

The commercial examples and broker economics were discussed in a report by Finance Magnates. The FCA's July 2026 Mills Review considered how AI agents might move influence towards firms that own the customer interface. It warned that automated comparisons could weaken the loyalty brokers have traditionally built through their brands, as agents select products against criteria such as price, risk or suitability.

The FCA's concern is that firms could need to compete for the attention of software agents as well as consumers. If assistants become the route through which people discover and select investments, brokers risk being reduced to execution and custody providers. In that case, the broker may carry the account, integration and control responsibilities while another company shapes the customer's choices.

Research into connected agents has also exposed practical uncertainty. Condor Capital Wealth Management's Q2 2026 report on digital advice examined AI agents linked to funded brokerage accounts. It described tools with useful capabilities, alongside setup friction and open questions about reliability, security and liability. The researchers cautioned that their testing period was short and should not be treated as a conclusive forecast of future results.

A Bank of England and FCA survey in October 2026 found that 86% of responding financial services firms used AI, while 7% of reported systems were considered highly material to their businesses. The survey covered financial services broadly, not retail brokers specifically, so it indicates a difference between adoption and material business impact rather than proving that brokers are earning returns from AI.

What this changes for operators

For any agent-connected account, teams should establish which account data an assistant can access, which order types it can submit and how client-defined parameters are enforced. The source does not specify the integration or control design behind the examples. Operators should ask how authentication, monitoring, exception handling and compliance review will work across the broker and the AI provider.

Teams should also track service costs and customer outcomes separately from usage. More queries, visits or trades do not by themselves show better retention or stronger economics. Before assigning an improvement to AI, compare users with appropriate controls for their prior engagement and technical sophistication, and include the cost of models, integration, monitoring and compliance in the assessment.

Operators should also establish which party owns the customer interface and how the broker's account services depend on the AI provider's access. The source does not detail the commercial or technical terms of these connections. Those are questions to resolve before the broker commits to an operating model, particularly if the agent becomes the route through which customers research and place trades.