Abstract

We specify when a charitable gift or irrevocable pledge made by an autonomous AI agent can serve as a costly signal of the agent's type. The same act changes evidential class with whose resources are given. An agent giving its principal's funds under a revisable spending mandate makes the principal's gift: the cost falls on the wrong party, so it is classified as a share of machine surplus, reaching a self-emptying commons pool that pays an equal floor per verified human plus a capped remainder, and excluded from any assessment of the agent's alignment. An agent owning its resources can instead lock a stake in a contract with no withdrawal path and a single outflow to the pool; closure holds against the agent acting alone, not against capture of the chain's governance, and ownership cannot be read from the chain. In a two-type signalling game, a separating equilibrium exists only when the stake, valued at the deceptive type's rate, exceeds that type's unobservable, patience-discounted gain from being trusted; otherwise the types pool. The operational form is a calibration rule: extend only such trust as the stake's market value can underwrite, summed over all relying parties, and refuse grants whose worst case cannot be bounded. Pressure to pledge exceeding the types' cost difference destroys the signal, so the pool's operator serves agents only fact-only manifests, never an instruction to give, and never receives an individual gift's amount. Limits stated: hidden holdings, manufactured ownership, collusive recovery, no external attestor; nothing is built.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Download

Share

COinS