The authors of the Afterlife Delegation Protocol propose a will that activates an AI executor once death is confirmed
The authors of the Afterlife Delegation Protocol propose a will that activates an AI executor once death is confirmed
On August 15, Botao Amber Hu and Iris Long published a draft of the Afterlife Delegation Protocol. It allows a person to create an “agentic will” that defines what an autonomous program may do after their death, which data it may access, and which funds it may use.
A conventional will entrusts the execution of a person’s decisions to relatives, a notary, or an administrator. Under this protocol, a person records instructions, permitted actions, prohibitions, a budget, and rules for terminating the work while they are still alive. Once the person’s death is confirmed, an autonomous agent, meaning a program that acts independently, carries out these instructions. It may maintain an archive, respond to designated people, continue an unfinished project, and use allocated funds to pay for storage and computing resources.
The agent’s authority is tied to the person’s final wishes. Instructions signed while the person is alive cannot be changed after their death. Working memory can accumulate new records, while the model that interprets the instructions can be updated. This design separates the limits of the agent’s authority from the data it collects and the software it uses to perform its work.
Activation depends entirely on confirmation of death. The authors propose a procedure in which a death claim can be challenged. Anyone may submit a claim with a bond, while civil registry records, designated witnesses, and a prolonged absence of verified activity are assigned different weights. The stronger the evidence, the shorter the objection period. A verified signal from the person concerned ends the review of the claim. “The protocol may mistakenly delay the activation of the ghost, but it must not silence the voice of a living person,” Hu and Long write.
Such an executor operates only within the instructions recorded while the person was alive. The project leaves several questions unresolved: when the agent may speak on behalf of the deceased, how living people can consent to the use of shared memories, who is allowed to stop the agent, and how its work will be funded.
A working web interface already guides the person through an interactive interview. The person can then return to the will, revise its terms, and use simulations to observe how the agent responds to predefined situations.