Abstract

This disclosure describes a lifecycle that every certified AI system follows, joined into one cycle that its owner can see and the public can check. The cycle has five stages: start, check, run, govern, and stop and restore. At every start, before any capability is unlocked, the system verifies its own integrity and its owner's key, and then presents a start statement: a short, plain-language account of what it is permitted to do in this session, under which safety profile, for how long, and how to stop it, with the owner's stop shown first. Running cannot begin until the statement has been presented and, where the profile requires, acknowledged. Before running, credentials and inputs are checked against signed proof, and where no proof exists the absence is stated rather than guessed. While running, a governor keeps the system within its declared limits and steps it down in grades when needed. The owner's stop works through its own path and never depends on the start, check, run or govern stages working correctly. After a stop, the system cannot simply be switched back on: it is repaired and hardened offline under the owner's authority, the repair is recorded, and the system returns only through a fresh verified start, with a new start statement that names what changed. Every start statement, check, limit, step-down, stop, repair and restart is signed and written to one registry. The same registry serves the system's own governance and a public lookup, so that anyone can verify, within privacy limits, the governance record of a certified system, and an operator cannot keep one record for regulators and another for the public. The same start statement is used when people, rather than machines, join a certified programme, so that the cycle is recognisable to everyone it affects.

Creative Commons License

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

Download

Share

COinS