Miden co-founder
@azeemk argues the answer to rogue AI transactions may be simple: don’t ask the agent to behave, make the forbidden transaction impossible.
"We ended up creating a solution to our own problem, which we call Guardian. You can think of it like a two-of-three multisig, where you have a hot wallet, a cold wallet key that's stored, and then you can have a third guardian that would be a signer."
"As we started building out, we started seeing, wait, you can actually put an AI in here and give the AI parameters it needs to follow."
"The human would be able to enable guardrails into Miden Guardian, and then have it do the things it wants while being comfortable knowing you can't somehow convince the agent to send $300,000 to someone."
"But if you could encode that it literally can't, that's fascinating."
@0xMiden