Binance launches Agent OS, a developer layer that enables AI applications to read market data, inspect designated subaccount information, and place trades through Binance infrastructure.

Binance uses the Model Context Protocol to let AI apps discover and invoke external tools for market data and trading functions without bespoke interfaces for each AI product.

The platform supports multiple AI tools—ChatGPT, Claude Code, Codex, Cursor, and other independent agents—and lets an agent be assigned to a dedicated subaccount with configurable permissions and revocation controls.

A dedicated subaccount contains an agent’s funds, positions, and history separately from other trading activity, while the agent can view balances and portfolio data but cannot access non-trading personal information.

Permissions can restrict products and functions available to the agent, and Binance monitors orders, with the agent’s broader reasoning and external data sources remaining inside the AI app, potentially leaving validation gaps.

Binance positions Agent OS as infrastructure for developers and quantitative teams rather than a provider of investment advice, while noting that AI-driven trading could raise regulatory and risk considerations.

In addition to market data access, the system enables the agent to place trades through Binance, leveraging the platform’s underlying trading infrastructure.

Details on maximum order sizes, default loss limits, cooling-off controls, and compensation mechanisms for agent-generated trades are not disclosed, raising questions about accountability in case of misinterpretation or faulty data.