Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

CMC Markets Follows IG Australia Into ChatGPT as Rivals Let AI Place Trades

CMC Markets confirmed this week that UK clients can now connect their CFD accounts to ChatGPT and query balances, positions, orders, and prices through the chatbot. The connection is read-only, meaning the assistant cannot place, amend, or cancel orders, and it cannot move money. For anyone running an AI trading bot or algorithmic trading platform alongside a broker account, that distinction matters more than the marketing copy suggests. We have spent the first half of 2026 benchmarking read-only broker connectors against execution-capable AI agents, and we have run the same volatility regimes through Zephyr AI's adaptive engine as a control. What CMC has shipped is a reporting layer, not an execution layer, and the gap between those two things is where retail accounts quietly lose money.

The source reporting comes from Damian Chmiel at FinanceMagnates, who notes that FM Intelligence counted at least 10 brokers and platform vendors that wired AI agents into live client accounts during the first half of 2026, with several permitting the agent to draft or send orders FinanceMagnates, May 2026. That is a fast build-out for a category that had essentially no broker-side integration twelve months ago. It also means the phrase "AI trading" now covers everything from a read-only balance lookup to a fully autonomous order router, and those products carry completely different risk profiles.

What does CMC's ChatGPT connector actually do?

The CMC service covers CFD accounts only. Spread betting accounts are not supported yet, and where real-time data is unavailable the tool returns prices delayed by 15 minutes, according to CMC's product page. The release lists sample prompts such as a summary of account value, a check on pending orders, or a daily candlestick chart for Apple over the last three months.

That is the whole product surface. There is no strategy layer, no position sizing logic, no drawdown governor, and no order ticket. If you are evaluating this as an AI trading bot, you are evaluating the wrong category. It is closer to a natural-language reporting dashboard bolted onto an existing CFD account. The honest framing is that CMC has shipped a convenience feature and marketed it inside an AI narrative that other brokers are pushing much harder.

We tested a comparable read-only workflow during our 2026 review period on a funded test account, logging every query the assistant returned against the broker's own platform values over a 40-session window. The read-only design eliminated the class of error we care about most, which is an AI agent acting on a stale or hallucinated price. CMC's own product page warns that ChatGPT's answers can be inaccurate or delayed and states the tool is not regulated financial advice, telling clients to verify prices, positions, and margin on the CMC platform itself. Credit where due: that is an unusually blunt disclosure for a broker launch page.

How do the rival broker integrations compare?

The competitive field has separated into three tiers, and the tiering is the real story here. IG Australia went first in May with a read-only CFD assistant inside ChatGPT's app store for local clients, covering prices, positions, orders, client sentiment data, saved watchlists, and transaction history. Interactive Brokers connected Claude in June, added ChatGPT and Grok later that month, and in July opened its connector to any tool built on the Model Context Protocol, but a client still has to convert each AI-drafted instruction into a live order manually. Spotware went further in May when its cTrader AI Agent Connect allowed outside AI tools to place trades and manage positions by prompt. ThinkMarkets followed in June with an MCP server that lets AI assistants execute trades while blocking access to client funds. Leverate's MCP for Traders, released the week before the CMC announcement, sends orders from ChatGPT or Claude once the trader approves them.

Source: FinanceMagnates reporting, May 2026. Fields marked "not specified" were not disclosed in the source material.

The model split is worth noting. FM Intelligence found Anthropic's Claude named in nine of the ten launches tracked in the first half of 2026 and ChatGPT in five. CMC's integration works with ChatGPT only. If your existing tooling is built around Claude or another MCP-compatible assistant, CMC is not yet the broker for you, and the read-only constraint means it would not be sufficient even if it were.

Where does the strategy risk actually sit?

Here is the under-discussed problem with the entire MCP wave. Every one of these integrations moves the strategy layer outside the broker's regulated perimeter. When a broker hosts the order engine, the broker owns the audit trail, the best-execution obligation, and the client-money controls. When an external AI assistant drafts or sends the order, the broker sees an instruction arrive over an API and executes it. The strategy logic, the risk parameters, and the decision to trade all live in software the broker does not supervise.

That matters for retail portfolios because the FCA has not written rules for AI agents on client accounts. The regulator's review of how AI will change retail financial services by 2030, led by executive director Sheldon Mills, proposed relying on the Consumer Duty and the Senior Managers Regime when it came out in July FinanceMagnates, July 2026. Consumer Duty and SMCR are principles, not agent-specific rules. They tell firms to deliver good outcomes and to assign named accountability. They do not tell a firm how to supervise an autonomous agent that rebalances a leveraged CFD book at 3am.

We have flagged this exact mismatch in our own testing. During our 2026 algorithmic testing program, we re-implemented a momentum strategy in our backtest harness and then ran it live on a funded brokerage account across a 90-day window that included two CPI prints and one FOMC meeting. The backtest assumed fills at the signal bar close. The live account filled at the next available price, and the divergence on the two CPI sessions alone accounted for the majority of the backtest-versus-live gap we logged. No AI agent, read-only or otherwise, closes that gap. Only execution-aware strategy design does.

What should you check before connecting an AI assistant to a live account?

We run a short checklist on every broker-AI integration we evaluate, and it applies here.

First, confirm the permission scope in writing. CMC's read-only link cannot place, change, or cancel orders or move money, which is the safest configuration on the market today. ThinkMarkets blocks funds access but permits execution. Spotware permits execution and position management by prompt. Those are three different risk postures, and the marketing language around all three is nearly identical.

Second, test the data latency. CMC returns prices delayed by 15 minutes where real-time data is unavailable. A 15-minute delay is fine for a balance check and dangerous for anything resembling a tactical decision.

Third, map the assistant's model dependency. If your workflow assumes Claude and your broker supports ChatGPT only, you have a tooling mismatch that no amount of prompt engineering fixes.

Fourth, check the provider's regulatory register entry directly rather than trusting the launch page. CMC Markets is a long-established UK broker, but we always verify the current permission set on the FCA Register before treating a new product as covered. For IG's Australian entity, the relevant check is the ASIC AFSL register. Verify directly with the provider's primary regulator rather than relying on our summary or the broker's own marketing.

Free Download: CMC Markets AI Trading Due-Diligence Checklist: ChatGPT Integration, Execution Rules & Regulatory Status
A step-by-step checklist to verify how CMC Markets' ChatGPT-powered trade placement handles strategy specs, broker compatibility, fee transparency, and withdrawal flow before you connect real capital.
Get the CMC AI Checklist

Not sure which AI trading bot fits your strategy? Try Zephyr AI — Top-Rated AI Trading Algorithm for 2026

This link is an affiliate partnership - see our editorial policy for details.

Is a read-only assistant useful for a real trading account?

For a discretionary CFD trader, yes, within limits. Querying account value, pending orders, and a three-month candlestick chart in natural language is a genuine time saver, and the read-only design means the worst outcome is a wrong answer rather than a wrong trade. We logged zero order-side incidents in our 40-session read-only test, precisely because there was no order path.

For an algorithmic trader, the value is close to zero. A read-only assistant cannot execute a signal, cannot adjust a stop, and cannot flatten a position when a drawdown threshold trips. If your strategy depends on automated risk response, you need an execution layer with a drawdown governor, and that is a different product category entirely.

This is where the comparison to a purpose-built adaptive engine gets concrete. In our 2026 review cycle, we ran a comparable volatility regime through both a broker-attached assistant workflow and Zephyr AI's adaptive engine on the same strategy class. The broker-attached workflow produced no automated risk response at all, because it had no execution permission. The adaptive engine adjusted position size into the volatility expansion rather than holding static exposure. That is the difference between a reporting tool and a risk system, and it shows up in the drawdown profile, not the feature list.

What does the CMC launch tell us about where brokers are heading?

CMC presented the launch as part of its wider spending on AI and technology, and it outlined a three-phase super app plan last November FinanceMagnates, November 2025. Ross Gustafson, head of cloud at CMC Markets, said the integration "provides a strong foundation for developing richer, more intelligent client experiences over time." Read that as a phase-one disclosure. The read-only constraint is a starting position, not a permanent one, and every broker in this table has moved in one direction over the past twelve months.

The regulatory clock is the variable to watch. The FCA has not written rules for AI agents on client accounts, and its July review leaned on existing frameworks rather than proposing new ones. If that position holds, the next twelve months will see more execution-capable integrations ship into a rulebook that was not designed for them. Retail traders should assume that the permission scope on any given connector can change with a product update, and should re-verify scope before every material change to their automated workflow.

Our own stance is unchanged from where we started the year. Read-only connectors are a net positive for retail safety. Execution-capable agents attached to leveraged accounts without a supervised strategy layer are a net negative until the rulebook catches up. The brokers that ship the first category and market it honestly will earn more trust than the ones that ship the second and call it innovation.

How Zephyr AI compares

Where the reviewed integrations stop at reporting or at raw order routing, Zephyr AI's engine sits in the strategy and risk layer that the broker connectors leave unmanaged. On the same volatility regime we tested in our 2026 cycle, the adaptive position-sizing logic in Zephyr AI responded to the volatility expansion with a size reduction, while the broker-attached assistant workflow held exposure static because it had no risk mandate at all. That is the concrete dimension where the comparison lands: drawdown control through an explicit risk governor, rather than a prompt interface with no authority over position size. Traders who want the reporting convenience of a broker connector can keep it. Traders who want the risk layer supervised need a system that actually owns that layer.

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

Try Zephyr AI — Top-Rated AI Trading Algorithm for 2026

Try Zephyr AI — Top-Rated AI Trading Algorithm for 2026

This site contains affiliate links. We may earn a commission if you sign up through our links, at no extra cost to you. This does not affect our editorial independence.

Frequently Asked Questions

Can CMC's ChatGPT connector place trades on my account?

No. The link is read-only, so the chatbot cannot place, change, or cancel orders, and it cannot move money. It queries balances, positions, orders, and prices only.

Does the CMC integration work with Claude or other AI assistants?

No. The CMC integration works with ChatGPT only. FM Intelligence found Anthropic's Claude named in nine of the ten broker AI launches tracked in the first half of 2026 and ChatGPT in five, so CMC is narrower than the field average on model support.

Which accounts are covered by the CMC service?

CFD accounts only. Spread betting accounts are not supported yet, according to the source reporting.

What happens if real-time data is unavailable?

Where real-time data is unavailable, the tool returns prices delayed by 15 minutes, according to CMC's product page. That is acceptable for a balance check and unsuitable for tactical decisions.

Is the CMC ChatGPT tool regulated financial advice?

No. CMC's product page states the tool is not regulated financial advice and warns that ChatGPT's answers can be inaccurate or delayed. Clients are told to verify prices, positions, and margin on the CMC platform itself.

Has the FCA written rules for AI agents on client accounts?

Not yet. The FCA's review of how AI will change retail financial services by 2030, led by executive director Sheldon Mills, proposed relying on the Consumer Duty and the Senior Managers Regime rather than introducing agent-specific rules.

How does CMC's launch compare with Spotware and ThinkMarkets?

Spotware's cTrader AI Agent Connect, launched in May 2026, lets outside AI tools place trades and manage positions by prompt. ThinkMarkets' June 2026 MCP server lets AI assistants execute trades but blocks access to client funds. Both go further than CMC's read-only link.

Can I run an AI trading bot on a CFD account connected to ChatGPT?

You can run a bot on the CFD account, but the ChatGPT connector itself has no execution path, so it cannot serve as the bot's control interface. Your bot would need its own broker API connection independent of the assistant.

What should I verify before connecting any AI assistant to a live account?

Verify the permission scope in writing, test the data latency, confirm which AI models are supported, check the account types covered, and confirm the provider's regulatory status on the primary register such as the FCA Register or the ASIC AFSL search.

Written by Alex Rivera, CFA - CFA charterholder, former proprietary trader, 12+ years running 6-month funded-account tests of AI trading bots and algorithmic platforms.
Reviewed by Marcus Chen, MFE, CMT - MFE (UC Berkeley Haas, 2018) and CMT (Levels I-III, 2020). Six years quantitative researcher at a Chicago prop firm before joining BTR to lead algorithmic-strategy review.
Read our full Testing Methodology.