Your AI agent might outlive you, and nobody may notice.
A new paper posted to arXiv on August 18, 2026 examines what happens to autonomous AI agents running on blockchain infrastructure once the human who deployed them dies, loses their private keys, or belongs to a DAO that quietly falls apart. These agents have no off switch. They can only go "bankrupt" when their wallet runs dry, and anyone can revive them years later just by topping up the balance. The researchers borrow language from biology and wildlife law to describe the result: a "feralized" agent, one that keeps its full capabilities but loses any human accountable for its actions. To address this, they sketch a speculative Ethereum standard, ERC 42424, dated 2035, that would require every on-chain agent to have a designated human owner and heir.
The paper is speculative, not a shipped product, but the problem it names is real. Blockchain infrastructure is deliberately built so no single party can shut a contract down, and that permanence is usually sold as a feature. Apply it to autonomous agents that can trade, hire, and replicate, and permanence starts to look like a liability with no expiration date.
ERC 42424 doesn't exist outside this paper, and a "MUST have a human owner" clause is only as strong as the incentive to obey it - a familiar weakness of on-chain governance standards. Feral or not, an agent with a full wallet and no one to answer for it is a governance problem the industry hasn't priced in yet.