Bybit has launched AI Sub-Accounts, a new account type built for traders connecting AI agents to live exchange infrastructure.
The feature is now live for all Bybit users and separates AI-controlled activity from a user’s main account. The structure gives traders a dedicated environment for automated strategies, with controls for asset caps, fund movement, leverage settings and trading permissions.
The launch targets one of the biggest risks in agentic trading: giving software too much control over a full exchange account. A compromised agent, broken script or badly configured prompt can create unwanted trades, transfer requests or liquidation risk if permissions are too broad.
Bybit’s setup keeps AI activity inside a ringfenced sub-account. Agent transactions remain confined to that account, while parent-account funds stay outside the agent’s direct control. Users can also set a maximum balance cap, with Bybit’s default cap placed at $5,000 unless the trader changes it.
The account is API-only. It does not allow normal login access or in-app account switching, which reduces the risk of manual account takeover through the AI-controlled profile. Traders can still monitor activity from the parent account while keeping execution permissions limited to the agent’s assigned environment.
API Controls Become The Product Layer
The AI Subaccount setup lets users create the account, connect an AI assistant through public-key and API-key steps, then manage permissions from the main account. Transfer-in requests, transfer-out permissions, movement between trading and funding accounts, margin leverage and contract leverage can all be scoped separately.
That permission model matters as AI trading moves from market summaries into execution. Bybit already supports AI-assisted trading through its AI Hub, where compatible assistants can query markets, manage positions and place trades across exchange functions. Sub-accounts add a stronger control layer around that workflow.
Crypto traders have already seen how execution settings can turn small mistakes into large losses, including a Hyperliquid TWAP order error that turned a position adjustment into a much larger trade. AI agents raise the same issue in a new form because the system may interpret instructions, call APIs and react to market data faster than a user can manually intervene.
For automated strategies, the useful controls are not cosmetic. Separate balances, no withdrawal permission, leverage caps, API scoping and live monitoring are now part of the trading stack. The same risk logic applies across machine-learning trading systems, where models can support execution but still need hard limits before real money is connected.
Bybit’s launch does not make AI trading safe by default. It gives users a narrower blast radius when an agent fails, gets compromised or executes the wrong instruction, which is becoming a core requirement as crypto exchanges move trading interfaces from dashboards into autonomous software.
Be the first to comment