Binance, the world’s largest cryptocurrency exchange with more than 300 million registered users, on Thursday launched a platform where AI agents can analyze markets and execute trades on behalf of users, bringing autonomous AI directly into its real money management business.
The platform, called Agent OS, allows developers to connect AI applications and agents to Binance’s financial infrastructure. This brings support for the newly introduced Model Context Protocol (MCP) in addition to the exchange’s existing tools and services such as the Binance API, Binance Wallet Agentic Hub, Binance x402 Transaction Validation and Payment Facilitator API, and Binance Skill Hub. The platform also integrates with tools such as OpenAI’s ChatGPT and Codex, Anthropic’s Claude Code, and Cursor, allowing users to authorize agents to access market data, view account information, and execute trades.
But as the AI race moves away from chatbots answering questions to agents who can take action, Binance is putting much of the onus in checking users, who must ultimately decide which agents can access and trade and set limits on what they can do.
“Instead of complete freedom, we put in the hands of users the power to have fine-grained access control over what they can do through their agents,” Jeff Lee, vice president of product at Binance, said in an interview. “To protect users’ funds, we place (controls) at the account level.”
Binance primarily does this through dedicated “subaccounts,” which users can assign to agents and configure for specific activities such as spot trading or futures trading. Withdrawals from these sub-accounts are blocked by default, creating a sandbox around agent activity, Lee told TechCrunch.
Users can also choose whether the AI agent needs approval for each order or can execute trades autonomously once permissions are set, a Binance representative said. Binance does not place individual limits on the amount that its AI agents can trade or lose, so the amount a user sends to a subaccount effectively acts as a limit.
When asked if Binance can figure out what factors prompted an agent to make a particular trade, Lee said that inference is done outside the system, either on the user’s computer or within an AI application of their choice. “We just don’t understand the basis of user behavior,” he says.
This means that while Binance can monitor the resulting trading activity of its agents, there is limited visibility into whether decisions were influenced by misinformation or manipulation.
When asked what happens if an agent is manipulated through a prompt injection attack or otherwise compromised, Lee again pointed to subaccounts as the main line of defense. Binance also stated that existing security, risk management, and anti-money laundering policies for the subaccount API will be applied to Agent OS upon release.
Trading is one of the first use cases Binance is targeting. Nevertheless, Lee said agents can monitor the market, conduct research and risk analysis, react to signals, and autonomously place orders and implement strategies such as arbitrage.
Agent OS is designed to connect agents to payments and on-chain activities. Binance’s x402 integration allows agents to send and settle payments, while Agentic Wallet allows agents to interact with tokens and decentralized finance protocols.
Unlike exchange trading, where Binance does not impose individual limits on the amount an agent can trade or lose within its subaccounts, Agentic Wallet trading has daily limits set by Binance. The company says regular swaps are limited to $50,000 per day, the default daily limit for DeFi transactions is $100,000, and x402 payments are capped at $20 per day.
Lee said AgentOS is Binance’s “first step” in providing developers with a platform to build AI-powered applications that work across crypto and traditional markets.
Binance is not the only company opening up its infrastructure to AI agents. Rival cryptocurrency exchanges are moving in the same direction, using MCP and other developer tools to give AI applications direct access to market data and trading systems.
In March, Kraken released an open-source command-line tool with a built-in MCP server that allows AI agents to perform actions such as spot and futures trading. In June, Coinbase released Coinbase for Agents, which connects AI agents directly to users’ accounts and allows them to perform transactions, payments, and other financial workflows within user-set limits. Similarly, OKX introduced the open source MCP toolkit earlier this year to enable agent trading on its platform.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
