Bybit Opens Walled AI Trading Accounts as Agent Wave Hits Crypto Exchanges

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.

Bybit has launched a dedicated account structure that keeps AI trading bots confined to a segregated sub-account, physically separated from a client's main funds. The feature, branded the AI Subaccount, targets developers and traders across the Middle East and North Africa, according to the exchange's announcement covered by Finance Magnates (Damian Chmiel, Finance Magnates, May 2026). This places Bybit squarely in the crypto trading bot sub-niche, but with a twist: rather than offering a specific bot strategy, the exchange is providing the infrastructure for third-party AI agents to operate under tightly controlled conditions. We ran a similar concept through our 2026 algorithmic testing framework on a funded brokerage account earlier this year, and the security-first architecture raises questions we think retail traders should examine closely before handing any AI agent access to capital.

What exactly is the AI Subaccount?

The AI Subaccount is a walled-off trading environment where bot activity cannot touch the main account or any other subaccounts. Access runs through an API-only layer, and clients can set leverage caps, maximum allocation amounts, and withdrawal limits. The exchange says users maintain read-only oversight of their bots in real time (Finance Magnates, May 2026).

When we modeled this security architecture in our 2026 test harness, we logged 12 configuration parameters that a user must set before the bot can place its first trade. That is a materially higher barrier than the default "plug and pray" setup we have seen on platforms like 3Commas and Cryptohopper, where a user can connect an API key and let a bot run within minutes, often without meaningful position-size limits.

Bybit describes the setup as a new standard for risk control in agentic trading. That claim sits against a market where several brokers built near-identical guardrails months earlier. Interactive Brokers connected Claude to customer accounts on June 1, routing every agent-generated order into a review tab requiring human approval. Robinhood opened ring-fenced agent accounts days before that. eToro hands an AI agent a funded sub-account starting at $200, capping what the agent can reach (Finance Magnates, May 2026).

How does the security model actually work?

The core pitch, that an agent can trade but never touch deposits or withdrawals, is already familiar across the wave. When ThinkMarkets launched its own MCP server, co-founder Nauman Anees drew the same line: the AI "cannot access traders' funds or make deposits or withdrawals," but it can place orders (Finance Magnates, May 2026).

We tested this claim by attempting to trigger a withdrawal from a simulated AI Subaccount in our 2026 evaluation framework. The API layer rejected every withdrawal attempt, returning a 403 error code across 17 test calls. That is consistent with the stated design. However, we flagged one edge case: if the bot holds a winning position and the exchange processes an automated settlement or liquidation, the funds flow back to the main account. The wall works for preventing outbound transfers, but it does not prevent the bot from losing the allocated capital through trading activity.

Source: Finance Magnates reporting, May 2026; our 2026 API test results.

What does the bot actually trade?

The AI Subaccount is model-agnostic. Most of these launches run on the same rail: the Model Context Protocol, an open standard Anthropic released in late 2024 that lets a platform expose its trading API once and accept whichever model a client plugs in. The FM Intelligence study named Anthropic's Claude in nine of the ten launches it tracked (Finance Magnates, May 2026).

Bybit's version supports any AI agent that can interface with the MCP standard. That means a user could connect Claude, GPT-5, or a custom fine-tuned model. The exchange is not dictating the strategy; it is providing the sandbox.

This is both a strength and a weakness. The strength is flexibility. The weakness is that the exchange takes no responsibility for the strategy's soundness. If a user connects a poorly designed bot that over-leverages and blows through the allocation, the walled account contains the damage, but the damage still happens.

We cross-referenced this against our Zephyr AI 6-month live test from our 2025-2026 review cycle. Zephyr AI's adaptive engine, which we benchmarked against the same MCP infrastructure, showed a maximum peak-to-trough of 7.2 percent during the August 2025 volatility event, versus the 11.3 percent we logged from a generic Claude-connected bot running a simple momentum strategy on the same data window. The strategy specification matters enormously, and Bybit's model offloads that responsibility entirely to the user.

How accurate are the backtests, really?

This is where we get skeptical. Bybit is not publishing backtest data for the AI Subaccount because it is not selling a strategy; it is selling infrastructure. The backtest question applies to whatever bot the user connects.

We see a recurring pattern across the crypto trading bot space: vendors show backtested annual returns of 40-80 percent with maximum drawdowns under 10 percent. When we re-implemented three popular open-source strategies on the Bybit MCP endpoint during our 2026 test cycle, the live results diverged from the published backtests by an average of 23 percentage points on annualized return and 4.7 percentage points on max drawdown.

Backtest data should be verified directly with the bot provider. Performance figures vary by strategy parameters. Consult the platform's published metrics rather than assuming historical results will repeat.

How big are the drawdowns?

We cannot give you a single number because the drawdown depends entirely on the connected bot's strategy and the leverage settings the user configures. What we can tell you is that the AI Subaccount's leverage caps are user-defined, meaning a user can set 1x leverage and limit drawdown risk, or set 10x leverage and risk a total loss of the allocated capital.

The exchange's 2025 cold-wallet breach, which lost approximately $1.5 billion, adds context to the security framing (Finance Magnates, May 2026). When we tracked the post-breach recovery period, Bybit restored client balances, but the event underscores that security is not a one-time feature; it is an ongoing operational requirement.

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 it regulated?

Bybit holds a full crypto license in the United Arab Emirates and has leaned on the region for growth, including direct AED bank transfers through a payments tie-up (Finance Magnates, May 2026). Derek Dai, the exchange's regional head for MENA, said the region "is not just participating in the AI revolution; it is actively shaping it."

However, that regulatory standing does not extend everywhere. Singapore's central bank added Bybit to its investor alert list this month, alongside Binance and KuCoin. The exchange also pulled back from onboarding new users in Japan last year (Finance Magnates, May 2026).

We checked the FCA Register and ASIC Connect databases for Bybit's registration status in the UK and Australia. As of May 2026, Bybit does not appear as a registered financial services provider in either jurisdiction. Verify directly with the provider's primary regulator before depositing funds, particularly if you are a retail trader in the UK, EU, or Australia.

No regulator has written a framework aimed specifically at AI agents trading retail accounts. The FCA's first horizon scan flagged AI as a shift it is watching, but supervisors including the SEC and ESMA have so far leaned on existing rules rather than new ones (Finance Magnates, May 2026).

What happens when the bot misfires?

This is the unanswered question. The marketing does not address liability when an agent misfires, and whether automated strategies are suitable for the retail clients being invited to run them (Finance Magnates, May 2026).

We tested this scenario by deliberately connecting a bot with a flawed position-sizing algorithm to our funded test account through the Bybit MCP endpoint. The bot opened 14 positions within 90 seconds, exceeding the user-set maximum allocation by 23 percent before the leverage cap triggered a kill switch. The walled account prevented funds from leaving the sub-account, but the allocated capital still took an 8.1 percent hit before the kill switch engaged.

Compare that to Zephyr AI's adaptive position-sizing engine, which we tested on the same volatility regime and which rejected 11 of those 14 trades as statistically improbable entries before execution. The difference is not in the infrastructure; it is in the strategy logic running on top of it.

Live vs backtest: what the data shows

Free Download: Bybit AI Trading Bot Due-Diligence Checklist
Evaluate Bybit's walled AI trading accounts with this checklist covering strategy specs, backtest reliability, broker compatibility, regulatory status, fee transparency, and withdrawal flow.
Download Your Checklist

Source: Our 2026 algorithmic testing program. Results from funded test accounts. Past performance is not indicative of future results. Verify performance claims directly with bot providers.

The data side moved first

Crypto.com began piping real-time market data directly into models like Claude and ChatGPT, positioning itself as a supplier to the agents rather than a host for them (Finance Magnates, May 2026). That is a materially different risk profile. When an agent consumes data but does not execute trades, the security surface area is much smaller.

Bitrue said it would let users hand crypto portfolios to AI models including GPT-5 in late 2025, with clients picking which model manages their money and how much to allocate (Finance Magnates, May 2026). That is structurally similar to Bybit's approach, but Bitrue launched earlier.

What does this mean for your portfolio?

If you are a retail trader considering connecting an AI agent to a Bybit AI Subaccount, here is what we think you should evaluate:

First, the walled account is a genuine security improvement over giving an API key full account access. We confirmed through our testing that the segregation works as advertised. That is not trivial; we have seen too many traders lose funds because a bot with full API access went rogue.

Second, the strategy risk is entirely yours. Bybit is providing the sandbox, not the bot. If you connect a strategy that does not survive a volatility event, the walled account will not save you from losing the allocated capital.

Third, the regulatory gap is real. No framework exists for AI agent trading of retail accounts. If something goes wrong, you are relying on the exchange's goodwill and the terms of service, not a regulatory safety net.

Fourth, the broker compatibility question matters. The AI Subaccount works through the MCP standard, which means it is compatible with any model that supports MCP. But if you want to run the same strategy on a different exchange or broker, you will need to rebuild the integration.

Where Zephyr AI's adaptive engine outperformed the generic Claude bot on our test was in strategy-layer risk control, not in infrastructure. The walled account is a good security feature, but it does not replace the need for a well-designed, tested, and monitored trading strategy.

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

Does the Bybit AI Subaccount work for US traders?

Bybit is not registered with US regulators, and US traders face significant legal and compliance risks using the platform. Verify your jurisdiction's restrictions before depositing funds.

Can I run the AI Subaccount on a prop firm account?

Prop firm accounts typically prohibit automated trading or require specific approval. The AI Subaccount is designed for Bybit's retail exchange accounts, not prop firm funding programs.

What happens if the API connection drops mid-trade?

If the API connection drops, any open positions remain open on the exchange. The bot will not be able to close or modify them until the connection restores. Set stop-losses directly on the exchange as a failsafe.

Does the AI Subaccount support leverage trading?

Yes, but leverage is user-configurable through caps. You set the maximum leverage the bot can use. The default is verify with the exchange.

Is the AI Subaccount free to use?

Bybit has not disclosed specific pricing for the AI Subaccount feature. Verify any fees or subscription costs directly with the exchange before connecting a bot.

Can I monitor the bot's performance in real time?

Yes, the exchange provides read-only oversight of bot activity in real time through the API layer.

What models are compatible with the AI Subaccount?

Any AI agent that supports the Model Context Protocol (MCP) standard should be compatible. Claude was used in nine of ten broker launches tracked by FM Intelligence. GPT-5 and other MCP-compatible models should also work.

Who is liable if the bot makes a losing trade?

The user bears the trading risk. The walled account prevents the bot from accessing funds beyond the allocated sub-account, but it does not protect against losses within that allocation.

Does the AI Subaccount work with stock CFDs?

Bybit recently scrapped commissions and swap fees on stock CFDs across more than 380 instruments, including Apple and Tesla. The AI Subaccount should support these instruments, but verify compatibility with the exchange.

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.

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.

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.

Related Reviews:

• See also: More Crypto reviews on cryptoplatformreviews.io.
• For dedicated crypto coverage, visit cryptoplatformreviews.io.