Prompt injection attacks on agent wallets are getting more sophisticated as LLMs do. With agent-authenticated wallets, agents can now hold a wallet and spend within limits their owner set, with those limits enforced by the wallet itself rather than promised by an app. A compromised agent can be cut off instantly, and until then it cannot act outside the limits its owner set. Co-authored by our very own @DonJohnsonSays(Virtuals Protocol)