Binance Now Lets AI Agents Trade Crypto for You — But You're in Charge of the Guardrails

What just launched

Binance, the biggest cryptocurrency exchange in the world with over 300 million registered users, has launched a new platform called Agent OS. The idea is simple to describe and a bit unnerving in practice: it lets AI tools — including OpenAI's ChatGPT and Codex, Anthropic's Claude Code, and the coding tool Cursor — look at crypto markets and place trades automatically on a person's behalf, without a human clicking "buy" or "sell" each time.

How Binance tries to keep this safe

Instead of just handing an AI agent full control of your main account, Binance requires you to set up a separate "subaccount" specifically for agent trading. Money you move into that subaccount is walled off from the rest of your funds, and withdrawals out of it are switched off by default — so even if an AI agent made every mistake possible, it couldn't directly drain your main wallet. You also get to decide whether the agent needs your approval before every single trade, or whether it can act on its own within limits you set.

There aren't separate trading limits beyond whatever amount you choose to put into that subaccount in the first place — so your own decision about how much to fund it is effectively your safety net. Separately, for actual money transfers and payments (as opposed to trades), Binance does enforce its own daily caps, like a $50,000 limit on regular transfers.

The part Binance can't fully see

Here's the catch: Binance can watch what trades an AI agent actually makes, but it can't see the reasoning behind those trades, because that thinking happens outside Binance's systems — on your own computer or inside whatever AI app you're using. That means if an agent gets tricked by bad information, or manipulated somehow, Binance would see the resulting trade but wouldn't necessarily know why it happened.

Not the only exchange doing this

Binance isn't alone here — rival exchanges Kraken, Coinbase, and OKX have all rolled out their own versions of AI-agent trading over the past several months, each using a similar approach: give the AI real trading power, but keep the safety limits in the user's hands rather than fully automating trust away.