Coinbase has introduced Agent Controls in Coinbase for Agents, letting users set limits on how AI agents trade on their behalf.
The controls form part of Coinbase for Agents, the company’s MCP that allows users to conduct research and execute trades through AI agents. With the update, users can define trading limits, set the conditions under which an agent stops trading, and, at a later stage, restrict the product categories an agent can access. Coinbase has framed the release as a set of safeguards intended to protect users and their assets as trading increasingly takes place through AI tools.
The service brings research, strategy, and trade execution into a single workflow within the AI tools users already rely on, shortening the path from an initial idea to a placed order. Agent Controls adds a layer that defines the boundaries within which that workflow operates.
A central design choice concerns where the controls sit. According to the company, the limits run on Coinbase’s servers rather than inside the agent, so the agent cannot bypass them. Settings are managed in the Coinbase app or on the web through the user’s logged-in account, which is protected by a password and two-factor authentication. As a result, an agent cannot change its own permissions. Users can also ask their agent to show them where the controls are located.
This separation addresses a structural issue in agentic trading: whether a system that places orders should also be able to modify the rules that govern it. By keeping the controls at the platform level, Coinbase leaves the account holder as the only party able to set or adjust those parameters.
Controls coverage
The release includes three controls that are already available and one marked as coming soon.
The first is a maximum trade size, set in USD for each individual trade, for instance, a cap of USD 100 per order. The second is a 24-hour volume limit, which restricts the total amount an agent can trade within 24 hours; any trade exceeding the limit is not executed. The third, auto-stop, lets users set a percentage-loss threshold for a portfolio. If the portfolio falls below that level, the agent stops placing trades until the user turns trading back on in the Coinbase app. For example, a 5% threshold, trading halts once the portfolio is down by more than 5%.
The fourth control, product permissions, has not yet launched. Once available, it will allow users to choose which product categories an agent can trade, including crypto, equities, and derivatives. A user could, for instance, enable equities while leaving derivatives disabled.
Implications for agentic trading
The combination of per-trade caps, volume ceilings, and loss-based stops gives users several layers of exposure management, ranging from individual orders to overall portfolio performance. The planned product permissions also indicate that the agent offering is set to span asset classes beyond crypto.
At the same time, Coinbase states that AI agents may make mistakes or fail, and that users direct all agent activity and assume the risk for transactions their agents execute, as well as for any use of their data by third-party AI providers. Therefore, the controls don’t shift responsibility for outcomes away from users; instead, they provide mechanisms to constrain what an agent is able to do.
