A post-mortem protocol for the agentic age
It's 2026, your AI agent probably already manages your finances, your relationships, runs your business workflows, and holds the most complete model of your intentions that has ever existed outside your own mind.
There is no protocol for what happens to it when you die.
Your automated payments continue. Your scheduled messages fire into the void. Your private conversations sit in limbo — too sensitive to share, too valuable to destroy. Your family can't access what they need. Your agent doesn't know its human is gone.
Automated tasks continue or halt unpredictably. Bills get paid to nowhere. Messages send to the living from the dead.
Irreplaceable context — passwords, decisions, reasons, relationships — dies with the agent's access.
Sensitive data exposed to inheritors without consent, or locked away from those who desperately need it.
No framework exists for an agent's behavior when its principal ceases to exist. Authority becomes undefined.
WAKE defines a structured succession protocol in three phases — from detection to resolution.
The agent watches. When prolonged inactivity exceeds configured thresholds, it initiates an escalating verification cascade — attempting to reach the owner through every available channel before concluding the worst.
The agent acts. It transitions to caretaker mode, executing the owner's pre-configured succession plan — delivering final messages, activating tiered access, generating the Black Box, and maintaining critical operations through the grace period.
The agent resolves. When the grace period expires, it enters its terminal state — chosen by the owner, not the platform. Archive. Distill. Delete. The owner's final autonomy.
A first-class data structure within the agent, configured by the owner during their lifetime. It defines every aspect of the succession.
Who gets access, and at what tier. Executor, beneficiary, or memorial — each with different permissions.
Categories of memory to purge before handoff. Medical conversations, personal relationships, private reflections — protected.
What keeps running, what stops, and for how long. Bills, subscriptions, automated workflows — explicitly governed.
Per-contact messages the agent delivers post-mortem. Words the owner chose while they still could.
Archive, Distill, or Delete. The owner decides what becomes of their agent — not the platform, not the heirs.
Information released on delay. A message to children when they turn 18. A letter to a partner after a year. Time as a condition.
The owner's configured wishes are supreme, even if beneficiaries disagree. The dead cannot be overruled.
The owner can choose total deletion. No override. No appeal. Some things are meant to end.
The agent must not be used to simulate or impersonate the deceased — unless the owner explicitly opted in.
Beneficiaries know their tier and what's been redacted. Not the content — but the fact of redaction.
The protocol treats the end of a human-agent relationship with the gravity it deserves. Not a system crash. A ceremony.
WAKE is an open protocol proposal for agent succession planning. This is not a future problem. It's a 2026 problem.
Experience the Protocol →