eToro connects eligible UK accounts to ChatGPT, Claude and Grok

eToro has switched on connectors that let eligible customers link their investment account to ChatGPT, Claude and Grok, bringing live portfolio data into a general-purpose AI chat window. The connectors, announced on 8 October 2026, use an open standard called the Model Context Protocol (MCP), which lets an AI application call tools and pull information from an external service such as a brokerage.

For UK customers the headline change is access, not advice: once a customer authorises the connection, the assistant can see real account details — balances, holdings and positions, not just public market commentary — and in some setups can propose a trade for the customer to approve. eToro's launch notice does not list the UK among its excluded markets, but it also says capabilities and availability vary by jurisdiction. A UK customer therefore needs to confirm two separate things: first, whether the connector is actually offered for their account and chosen assistant, and only then, once a connection exists, what the authorisation screen lets them permit in terms of data and actions.

This matters now because it moves the AI assistant from a research tool a customer manually briefs into something that can see — and in limited cases act on — a live, loss-bearing portfolio. eToro, Anthropic, OpenAI and xAI each carry their own rules on what the assistant can do and what happens to the data afterwards, and the UK's Financial Conduct Authority (FCA) has separately warned that AI-generated investment output is not held to the same standard as regulated advice.

How the connection works

eToro says the connectors sign customers in using their existing eToro credentials rather than requiring them to copy and paste an API key. After authorisation, the assistant can be granted read-only access, or — where the assistant and jurisdiction support it — read-and-act access. The categories of account information that can become visible include profile details, balances, holdings, performance, and open and historical positions and orders, but exactly what is shared depends on the permissions a customer grants on the authorisation screen itself, not on a single fixed list.

eToro says it does not receive the prompts a customer types into the assistant, or the assistant's replies. What eToro receives are instructions the assistant submits through eToro's own API, once a customer has approved them.

What each assistant can do

Capability is not the same across Claude, ChatGPT and Grok, and it depends on the version, plan and setup route a customer uses.

eToro attributes the Claude web/desktop gap to differences in how the two Claude clients run tools, and notes that features may change.

For ChatGPT, eToro's own guide does not state which subscription tier is required. OpenAI's documentation, which governs ChatGPT generally rather than the eToro connector specifically, says custom MCP apps of this kind are web-only as of 12 October 2026 — so a UK customer using ChatGPT's desktop or mobile app cannot use this custom connector at all, regardless of plan. Among web users, the plan determines what the connector can do: OpenAI says Pro customers can connect MCP servers with read and fetch permissions only, while full read-and-write MCP support is limited to Business and Enterprise or Edu customers.

eToro's launch messaging describes the connectors as needing no code or API key. Its own ChatGPT setup guide, however, requires switching on Developer Mode and manually entering a remote MCP endpoint — a step with more friction than "no technical setup" implies.

Approval is a control, not a guarantee

Where write access is available, eToro says an action is proposed for the customer to review and approve, and that funding and withdrawals stay inside eToro rather than happening in the chat interface. OpenAI's general MCP documentation adds a qualifier worth noting: ChatGPT "may" ask for confirmation depending on permissions, context and the apparent impact of an action, rather than always doing so in every circumstance. How this resolves for every order type, on every assistant, was not independently verified for this article.

eToro's own warning is blunt: AI systems can perform poorly, act quickly and be hard to stop, and a submitted trade can differ from what a customer requested or saw on screen. eToro tells customers to check positions and fees inside their eToro account rather than relying on the chat transcript. Investment values can fall as well as rise, and capital invested through eToro is at risk.

Where portfolio data goes once it's shared

eToro's privacy policy, last updated in September 2026, says disclosure to an AI provider rests on customer consent and covers whatever categories appear on the authorisation screen. Once shared, the AI provider — OpenAI, Anthropic or xAI, depending on which assistant is connected — processes that data as an independent controller under its own terms and privacy policy. eToro's own privacy policy continues to govern eToro's collection and disclosure of the data up to the point of transfer; it is the provider's subsequent processing that sits outside eToro's policy.

Disconnecting an assistant stops eToro sending any further information, but it does not delete data already handed over. A customer who wants previously disclosed account information erased has to approach the AI provider directly, not eToro.

OpenAI's documentation adds a further wrinkle specific to ChatGPT: on Free, Plus, Go and Pro plans, information accessed through a connected app may be used to train OpenAI's models if the customer's "Improve the model for everyone" setting is switched on. On Business, Enterprise and Edu plans, connector information is not used for training by default. Comparable detail for how Anthropic or xAI treat retrieved portfolio data was not found in the material available for this article.

Security considerations

OpenAI's own guidance on Developer Mode and MCP apps warns that connecting to an untrusted MCP server can create security risks including prompt injection, where malicious content manipulates the assistant's behaviour, and says responsibility for vetting a custom connector sits with the user or, in a business setting, the workspace administrator. This is general MCP risk context rather than any suggestion that eToro's specific implementation has been found vulnerable — no independent security assessment of eToro's connector was identified.

The regulatory backdrop

The FCA has said general-purpose AI tools such as ChatGPT are not regulated in the same way as authorised financial advice, and that AI-generated information on its own does not carry Financial Services Compensation Scheme or Financial Ombudsman Service protection. The regulator has also warned that AI can return information that is incorrect or out of date, and that it cannot predict future investment performance — a limitation that live portfolio access does not remove, since seeing real balances does not stop a model misreading them or generating a flawed judgement from them.

An FCA survey of 666 UK respondents aged 18 to 40 who owned or were considering investments, conducted on 24 July 2026, found gaps between how much these younger investors trust AI and what they understand about its regulatory status. 56% said they trusted AI tools, but 44% mistakenly believed AI-generated financial information was regulated, and 32% wrongly believed FSCS or Financial Ombudsman Service redress would apply if AI advice went wrong. 38% thought it was acceptable to base an investment decision solely on AI output. On the more reassuring side, 73% knew AI could give inaccurate information and 86% said sources should be checked.

Separately, eToro has cited its own commissioned research — a 13-country Opinium survey of 11,000 retail investors run between 13 and 28 August 2026 — finding that 56% already use, or are open to using, AI tools to help pick or alter investments. That figure spans multiple countries and respondents did not need to be eToro customers, so it should not be read as a UK-specific result.

eToro launched an earlier, separate product called Agent Portfolios in March 2026, which let an independently configured AI agent trade inside a dedicated sub-portfolio using an API key; at that time, eToro said standard ChatGPT and Claude.ai chatbots were not supported. The 8 October 2026 connectors are a distinct, later product, so that March statement should not be read as a continuing limitation on this separate connector service.

What isn't yet established

Several practical questions remain open based on the material eToro and the AI providers have published so far. There is no public confirmation that every eToro UK retail account is eligible for these connectors, as opposed to the UK simply not being on the excluded list. A full, current breakdown of exactly which tools and permission scopes a UK customer sees — and whether products such as leveraged positions or cryptoassets carry separate restrictions — has not been published in a form comparable across all three assistants. How long OpenAI, Anthropic or xAI retain portfolio data pulled through the connector, and whether it can persist in chat memory or logs after a customer disconnects, also was not established.

What to watch

Customers considering this should check eToro's own connector guides and privacy policy for the current list of data categories and permissions before authorising access, since both are described as subject to change. The FCA's consumer guidance on using AI for investment research sets out the regulator's position on AI tools more broadly. Anyone unsure what an assistant has been granted access to should review the authorisation screen again, and verify any proposed trade, fee or position inside their eToro account rather than from the chat reply alone.