Bybit has rolled out a new framework called AI Sub-Accounts, designed to put a firewall between your main trading account and whatever your AI agent decides to do with your money.

The feature, introduced on May 20, forces all external AI agents and third-party trading tools into dedicated sub-accounts that are completely segregated from a user’s primary Bybit account. Users can set configurable limits on contract leverage and position sizes, meaning an AI bot can’t suddenly decide to go 100x long on a memecoin at 3 a.m. while you’re sleeping.

How the sandbox works

AI agents operating under this framework are restricted to working solely within their assigned sub-accounts. They cannot touch funds or positions in a user’s main account, and their API permissions are tightly scoped.

Users retain full control over the risk parameters. They can configure maximum leverage ratios, cap position sizes, and define exactly what each AI agent is allowed to do through granular API permissions. If the bot hits its limits, it stops.

The AI Sub-Accounts system plugs directly into Bybit’s existing AI Hub, which already provides infrastructure for connecting large language model agents and automated trading bots. The Hub offers tools for monitoring and managing these agents, and the sub-account layer adds a risk management dimension that was previously missing from the equation.

What this means for traders

For retail users experimenting with AI trading bots, the immediate benefit is peace of mind. You can allocate a specific amount of capital to a sub-account, let an AI agent trade within those boundaries, and know that your main holdings are untouched regardless of what happens.

For professional and institutional traders, the implications are more structural. Multi-agent setups become far more manageable when each agent operates in its own sandbox with individually defined risk parameters. If one strategy blows up, the others continue operating normally.

The leverage and position size limits are particularly important. Hard caps at the account level, rather than relying on the bot’s internal logic, provide a failsafe that doesn’t depend on the agent’s code being perfect.

By restricting API permissions at the sub-account level, the attack surface for compromised API keys shrinks significantly. If a third-party bot’s infrastructure gets hacked, the damage is limited to whatever capital was allocated to that specific sub-account. Your main account stays intact.