MEXC, Hacken Publish Guide to AI Trading Safeguards for Crypto

MEXC and Hacken published a safety guide for AI crypto trading Tuesday, Oct. 27, urging platforms to check trades outside the model as MEXC plans to expand its AI tools.

It recommends limiting each agent to the credentials and permissions it needs, and applying separate checks for order size and total exposure. The guide also calls for caps on leverage and losses, an emergency stop, activity logs and repeated security testing, including red teaming. It addresses safeguards for both users and platforms. The handbook says financial limits should be enforced separately from the AI model.

The guide warns that withdrawal restrictions alone may not be enough to limit losses: An agent with trading access could act on manipulated inputs, buy an illiquid asset at an inflated price or open a highly leveraged position. The handbook recommends reviewing information an agent receives and matching human approval to the risk of an action.

The handbook says Hacken's lab testing found that an instruction embedded in a software changelog led an agent to run a script without explicit approval. MEXC CEO Vugar Usi Zade said, “greater capability should come with stronger safeguards.”

A MEXC online survey received 3,708 valid responses during Sept. 14-20, 2026, with respondents recruited through its social media, app and website. It found 84.8% of respondents said they had used AI for crypto research or trading in the prior six months. Those results describe the respondent pool, not U.S. crypto traders as a whole.

In a hypothetical scenario, 55% said they would allow automatic trading if they received notifications and could pause or revoke access. The figure rose to 71.8% when the scenario added limits on funds, positions, leverage and losses, with trading stopping automatically at the loss threshold. These responses measure stated preferences, not observed trading behavior.

In September, MEXC announced plans to expand its AI tools toward strategy generation and user-confirmed execution. It also planned a command-line interface for compatible agents operating within user-authorized permissions. The handbook offers guidance for users and platforms; it does not announce a new trading feature or say MEXC has deployed the recommended controls across its systems. TokenPost previously covered MEXC's AI-use survey.