YZi Labs discussed what would be required to trust an AI agent with millions in trading capital. YZi Labs posted on X.

Sandeep Nailwal said language-based research must be paired with mathematical analysis and checks for hallucinations before an agent makes a trading decision.

He said significant technological work is still needed on the AI side before crypto companies can adopt it.

Peter Pan of 1kxnetwork said trusting agents with capital requires risk management and a way to recover losses when problems occur.

He said the firm already uses agents to run processes and monitor positions, while humans still execute most trades.

Pan added that there should be some way to insure against malfunctions and hallucinations, including a way to claim capital back if something goes wrong.