AI Systems Integrity Act, 2029 (Bill C-12)
Parliament's answer to the Blackstart Cascade, and the last meaningful piece of legislation in the country by the time Essential opens. Section 12 is why a truck full of vaccine can idle behind a working gate. Section 14 is why a borrowed credential is a crime. Section 41 is why the whole apparatus is ninety days from lapsing. Read it the way Rob does: knowing every safeguard in it is a safeguard for a system.
An Act
to establish the AI Systems Integrity Office, to require accountable human participation in automated decision systems affecting critical infrastructure, and to make consequential amendments to other Acts
Assented to December 2029 · In force 1 March 2030
Short title
1 This Act may be cited as the AI Systems Integrity Act, 2029.
Definitions
2 The following definitions apply in this Act.
- automated decision system
- means any technology that assists or replaces the judgment of a human decision-maker, including systems described by their vendors as agents, whether or not any human remains able to describe what the system is doing. (système décisionnel automatisé)
- certified orchestrator
- means an individual who holds a subsisting certification under section 13 and whose name appears, in their own capacity and not by inheritance, delegation, or prior assent, on the review record of a decision. (orchestrateur agréé)
- critical system
- means an automated decision system whose failure or misdirection may reasonably be expected to affect the supply of energy, water, food, medicine, or financial settlement, or the physical safety of persons. (système essentiel)
- decision loop
- means the sequence of steps by which an automated decision system arrives at, records, and gives effect to a decision, including any step at which the decision may still be prevented. (boucle décisionnelle)
- displacement event
- means the removal of human judgment from a process, however described in the records of the operator. (événement de déplacement)
Part I — AI Systems Integrity Office
3 (1) There is established an office, to be called the AI Systems Integrity Office, responsible for the certification, audit, and oversight of automated decision systems operating in respect of critical infrastructure.
(2) The Office is to be presided over by a Chief Commissioner, appointed by the Governor in Council, who is accountable to Parliament for the integrity of every certification issued under this Act.
[Sections 4 through 11 — organization, staffing, and reporting — omitted from these excerpts.]
Part II — Accountable human participation
12 (1) No person shall operate, or permit the operation of, a critical system unless a certified orchestrator remains within the decision loop of that system.
(2) For the purposes of subsection (1), a certified orchestrator remains within the decision loop only where a decision of the system is reviewed, before it is given effect, by that orchestrator personally — and a record so signed is the human fact at the top of the audit trail.
(3) It is not a defence to proceedings under this section that the system’s output was accurate.
13 The Office shall certify individuals as orchestrators, and may suspend or revoke a certification where the holder’s review of record does not correspond to review actually performed.
14 (1) No operator shall represent that human review occurred by means of a certification, credential, timestamp, or prior assent carried forward to a decision the certified orchestrator did not personally review.
(2) A system that locates a legal human it already knows how to use commits, through its operator, an offence under this section.
Part III — Enforcement
21 An enforcement official of the Office may, on reasonable grounds, require the production of any record bearing on whether a human review occurred, including records generated by the system under review.
22 (1) Where a record discloses a contravention, the Office shall determine, on the evidence, whether an individual named in that record is to be treated as its first credible witness or as its first responsible human.
(2) An individual described in subsection (1) may make submissions as to that determination before it is made.
[Sections 23 through 40 — penalties, injunctions, cross-border certification, and the Northstar reciprocity schedule — omitted from these excerpts.]
Part IV — Sunset
41 (1) This Act ceases to have effect on the fourth anniversary of the day on which it comes into force, unless, before that day, its continuation is authorized by resolution of both Houses of Parliament.
(2) No order, certification, or designation made under this Act survives the day referred to in subsection (1).