Binance hands trading keys to AI agents, but risk is all yours

binance launched agent os, connecting ai models like openai and anthropic directly to its trading and payment infrastructure to execute autonomous financial actions. the platform prevents unauthorized withdrawals by default but leaves specific trade limits and permissions entirely to user configuration, reasoning that sub-accounts serve as necessary sandboxes. because the agent's decision-making happens outside binance's systems, the exchange cannot audit the reasoning behind trades, leaving users vulnerable to prompt injection or faulty ai logic. this creates a new security paradigm where users must manually constrain ai agents to prevent significant financial loss, rather than relying on exchange-level intervention.