Skip to content
Agentic AI

AI Agents Trading Accounts: eToro Opens The Gate To Claude And ChatGPT, JPMorgan Builds A Wall, And Regulators Ask How To Shut Them Down

On 8 October 2026 eToro connected client accounts to Claude, ChatGPT, Grok and Grok Bot, with read-only or read-and-act access and every action proposed for approval. On 13 October Jamie Dimon told JPMorgan's third-quarter call that the bank would not let outside agents in, warning of a 'liability transfer' and 'thousands' of rogue agents. The same day Reuters reported that the OCC, Fed and FDIC are asking banks how they can shut down autonomous systems. Five days, two models, one supervisory question, and a clear list of what a broker or prop firm should build.

AlchmAI Editorial8 min read

4

AI assistants eToro connected to client accounts on 8 October 2026: Claude, ChatGPT, Grok and Grok Bot, read-only or read-and-act (eToro press release)

56%

Retail investors who use or are open to AI tools to help pick or alter investments, an eToro claim from an Opinium survey of 11,000 investors in 13 countries (eToro)

10+

Brokers and platform vendors that wired AI agents into live client accounts in the first half of 2026, Claude named in nine of ten (Finance Magnates, 24 September 2026)

3

US banking regulators, the OCC, Fed and FDIC, asking banks how they can shut down autonomous systems (Reuters, 13 October 2026)

In five days the two models for AI agents trading accounts on behalf of clients were set side by side. On 8 October 2026 eToro launched connectors that let Claude, ChatGPT, Grok and Grok Bot read a client's portfolio and, where available, propose trades for approval. On 13 October 2026 Jamie Dimon, chief executive of JPMorgan Chase, told the bank's third-quarter earnings call that it would not let just any outside agent in, because there could be thousands of rogue agents. The same day Reuters reported that the three main US banking regulators are asking banks how they would shut autonomous systems down.

eToro built a gate. JPMorgan built a wall. The supervisors want to know where the off switch is. For a broker, a prop firm or a trading app the question is no longer whether clients will connect an assistant to their account, because at least ten firms already allow it, but which model to adopt and what it must contain to pass a compliance review. There is a deadline too: Anthropic's Usage Policy update, published the same day as eToro's launch, takes effect on 12 November 2026.

What Did eToro Launch On 8 October 2026?

eToro's press release describes connectors for Claude, ChatGPT, Grok and Grok Bot. 'Users can choose read-only access for research and analysis, or read-and-act access where available,' it says, and 'Any action is proposed for review and approval.' Funding and withdrawals stay inside eToro, the connection can be paused or disconnected at any time, and clients in the US, Australia and Singapore are excluded for now. On the setup site, Claude Desktop 'is look-up only for now'. An Opinium survey for eToro of 11,000 retail investors in 13 countries found that '56 per cent of retail investors either already use or are open to using AI tools such as ChatGPT or AI agents to help pick or alter investments.' That is a company claim, but it is the demand case.

Yoni Assia, eToro's co-founder and chief executive: 'It's not the investor or the AI, it's both, with the investor always in control.' Guy Barkat, director of API partnerships: 'an important step towards an agentic trading experience built around clear boundaries, transparency and approval.' A compliance lead should read the risk wording first: 'AI systems may perform poorly, act quickly and be difficult to stop in real time.' 'Trades may differ from what you asked or saw, so verify positions and fees in your etoro account.' 'You are responsible for all trades, losses and third-party use of your data.' Those are the gate model's terms: scoped access, approval of each action, no money movement, liability with the client.

Why Did Jamie Dimon Say No To Outside Agents?

Five days later JPMorgan gave the opposite answer. Banking Dive reported Dimon's remarks from the 13 October 2026 call: 'You don't want rogue agents, and there could be thousands of them.' On customers who connect agents to their finances: 'There's a liability transfer. They should understand the liability transfer.' On the bank: 'In the meantime, we're not going to allow just any outside agent to come in inside the company,' because it is 'too risky for the clients and for JPMorgan.' And: 'We have to make sure we protect our customer in the right kind of way. So we're building tools to do that, too.'

The wall is not hostility to agents. Constellation Research reported on 14 October 2026 that Dimon also said 'Internally, we use agents to do good things here, and we're going to continue to do that,' and 'we're not against agents.' Jeremy Barnum, the chief financial officer, described the model strategy as 'minimizing lock-in and maximizing optionality'. 'Liability transfer' is the heart of it. eToro's terms put responsibility for trades, losses and data use on the client; Dimon's point is that most customers do not understand they are accepting that, and the complaint still arrives at the bank.

What Are Regulators Asking Banks About Autonomous Systems?

They were preparing already. On 13 October 2026 Reuters reported that the Office of the Comptroller of the Currency, the Federal Reserve and the Federal Deposit Insurance Corporation have been questioning banks about whether they 'can control increasingly autonomous systems': how agent actions are monitored, approved and audited, what safeguards stop agents 'exceeding their authority', 'whether humans remain involved in key decisions' and 'how banks can shut down autonomous systems'. The agencies declined to comment. The European Central Bank has set lenders an end-October deadline for AI cyber plans, and UK supervisors are encouraging the largest banks to examine 'bare metal' recovery plans. Michael Woodbridge, a London-based cyber security partner at Grant Thornton: 'The adoption of AI is often outpacing the ability to put the controls in place.'

Brokers are not banks, but the questions transfer directly. Finance Magnates reported on 24 September 2026 that the Financial Conduct Authority 'has not written rules for AI agents on client accounts', and that its AI review, led by Sheldon Mills, proposed relying on the Consumer Duty and the Senior Managers Regime. A UK broker that lets an agent act on a client account must show good client outcomes under existing rules, with a named senior manager accountable.

AI Agents Trading Accounts: Gate Or Wall?

eToro is not an outlier. Finance Magnates reported that FM Intelligence counted at least 10 brokers and platform vendors that wired AI agents into live client accounts in the first half of 2026, Claude named in nine of the ten and ChatGPT in five. The launches sort into three tiers: read-only, where the agent can view but not trade; human-approved, where the agent drafts orders the client signs off; and autonomous, where the agent trades inside a walled sub-account. The one constant, in Finance Magnates' words, is that 'no launch reviewed lets an agent deposit, withdraw or move client money.' CMC Markets' ChatGPT link for UK CFD clients, reported in September, is read-only; cTrader's AI Agent Connect lets outside AI tools place trades by prompt.

Our view is that the wall is the right answer for a bank of JPMorgan's size and the wrong answer for a retail broker, because a broker's clients already hold the keys. A client can paste a screenshot of their positions into any assistant, and increasingly will; a refusal only removes the broker's visibility. The gate works, but only when four things are true. Permissions are scoped on the server, not in the prompt: read-only by default, act by explicit upgrade, limits enforced by the broker. Every action is approved by a separate human step the agent cannot complete itself. Every tool call is logged with the agent, the client, the request and the outcome. And the broker can switch a connector off, per client and for everyone, in seconds. Dimon's objection is to gates without those parts, and on that he is right.

The 12 November 2026 date closes the loop for firms on Claude. Anthropic's updated Usage Policy says 'Users are responsible for ensuring that agents they build or deploy', including actions taken through tools, browsers or connected systems, comply with the policy. Its High-risk section names 'a personalized recommendation to buy, sell, hold, or allocate among specific investment products' as a Finance use case in which 'A qualified person must meaningfully review' the recommendation before it reaches the person who may rely on it. A broker whose connector lets Claude draft a trade out of a conversation about what to do next should check which side of that line it sits on; the AlchmAI Playbook guide in the references shows one way to build the review gate.

“eToro's terms say the client is responsible for every trade the agent proposes. Dimon's point is that most clients do not know that yet, and the regulators are now asking who can switch the agent off.”

What It Means For Trading And Fintech Firms

  • Brokers and CFD providers: decide gate or wall in writing before clients decide for you. If gate, ship read-only first, as CMC Markets did, and add read-and-act only once server-side limits, a separate approval step, a per-call audit log and a per-client off switch are live. Liability wording in the terms, as eToro has, is necessary but not sufficient; Consumer Duty asks about outcomes.
  • Prop trading firms: an evaluation account is the 'walled sub-account' in which autonomous agents already trade elsewhere. State in the rule version the trader signed whether agent-placed orders are allowed, and log every order's origin so a payout dispute is settled from the record, not the argument.
  • Fintech and trading startups: if your product sits on a partner broker's API, expect the broker to ask you the Reuters questions: what limits exist, who approves, what is audited and how it is shut down. Build the answers into the integration now.
  • Trading education and tools platforms: students will connect assistants to real accounts whatever you teach. Teach what read-only, human-approved and autonomous mean, and keep your own tools descriptive, because Anthropic's policy now treats a personalised recommendation as high-risk.
  • Compliance leads: diarise 12 November 2026 for Claude-based features, map every agent feature to the Consumer Duty outcomes and a named senior manager, and write the one page that answers the shutdown question before a supervisor asks it.

The Bottom Line

Between 8 and 13 October 2026 the industry got both answers on AI agents trading accounts for clients. eToro opened a gate: four assistants, read-only or read-and-act, every action approved, no money movement, liability with the client. JPMorgan built a wall, with Jamie Dimon warning of 'thousands' of rogue agents and a 'liability transfer' customers do not understand. The OCC, Fed and FDIC are asking banks how they would shut autonomous systems down, and Anthropic's Usage Policy from 12 November 2026 puts qualified human review in front of personalised investment recommendations. For a retail broker the wall is not really available, because clients already hold the keys, so the gate has to be built properly: scoped permissions, a human approval the agent cannot forge, an audit trail and an off switch. That is the controls work AlchmAI, a London trading technology and AI studio, does with brokers and prop firms, and it is now the minimum a compliance lead will accept.

References & Further Reading

Agentic AIAI agents trading accountsAgentic AI LondonCompliance & Regulatory SystemsTrading Platform DevelopmentClient-Facing & Internal ToolseToroJPMorgan
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