Alberta plans first AI-drafted law: a test case for the Alberta Whisky Act
Alberta is preparing to use artificial intelligence to help draft legislation for the first time. The test case: an Alberta Whisky Act that defines what can be labeled "Alberta whisky."
Dale Nally, Minister of Service Alberta and Red Tape Reduction, says the file is ideal for a controlled trial. It's process-heavy, fact-based, and correctable without risking health or safety.
Why whisky is a smart pilot
Nally's brief is to set standards covering grain inputs, water sources, distillation, and labeling. If the AI misses the mark, the government can adjust without major fallout.
He met with Technology Minister Nate Glubish and Justice Minister Mickey Amery to align on guardrails. Justice will review the draft before it goes anywhere, giving the government an off-ramp if needed.
Human oversight is non-negotiable
Jonathan Schaeffer, professor emeritus of computer science at the University of Alberta, calls the plan innovative-if safeguards hold. His point is simple: AI can draft, but humans must own the result.
Models are trained on global data. Without careful editing, they can inject ideas that don't fit Alberta law, Canadian constitutional constraints, or the province's policy intent.
What "Alberta whisky" could mean
The government is consulting industry now. To distillers like Hansen Distillery's CEO, Keenan Pascal, the definition should reflect Alberta's grains, water, climate, and production practices.
There's an export angle too. Canadian whisky already performs well internationally. A clear Alberta standard could strengthen the story abroad.
Where other governments stand
Several provinces and territories say they're not using AI to draft laws. British Columbia's attorney general office notes that legislative drafting remains a human-led legal function to ensure accuracy, consistency, constitutional compliance, and policy alignment.
Alberta's move could be a first in Canada-if it survives review and reaches the Order Paper this spring.
What this means for government teams and legislative writers
If your team is exploring AI for drafting, treat this as a blueprint, not a shortcut. The tech can accelerate structure and language, but policy, legal intent, and accountability stay with humans.
A practical workflow to pilot AI drafting
- Define scope and constraints: Purpose of the bill, statutory authority, exclusions, and deadlines.
- Assemble source material: Comparable statutes, regulations, style guides, case law triggers, technical standards, and stakeholder inputs.
- Engineer the prompt: Specify legislative voice, jurisdiction, definitions, prohibitions, enforcement, transition, and cross-references.
- Use retrieval: Feed the model with your approved corpus so outputs anchor to Alberta context, not generic internet text.
- Generate a skeleton: Preamble, interpretation, scope, standards, compliance, enforcement, penalties, regulation-making, review clause.
- Run legal review: Check constitutional fit, existing statute conflicts, delegated authorities, and readability.
- Red-team the draft: Search for ambiguities, loopholes, and unintended incentives. Test adverse scenarios.
- Document provenance: Record prompts, source docs, model version, reviewers, and decisions for auditability.
- Consult and iterate: Industry, consumer protection, trade, and enforcement bodies should stress-test the text.
- Finalize and monitor: Build a plan for guidance, communications, and post-enactment review.
Guardrails to put in place
- Accuracy and accountability: No AI-only approvals. A named human owner signs off every section.
- Style consistency: Force adherence to your legislative drafting manual and plain-language standards.
- Jurisdictional fit: Ensure compatibility with federal law and existing Alberta statutes.
- Security and privacy: Keep prompts and data off public tools. Use vetted environments and avoid sensitive inputs.
- Bias checks: Review definitions and thresholds for unintended market effects or barriers to entry.
- Procurement clarity: Cover IP ownership, logs, model versions, and service levels in contracts.
- Change control: Version drafts rigorously to avoid accidental drift in obligations or rights.
What success looks like
- The AI helps structure a clean, complete first draft with fewer iterations.
- Human reviewers catch and fix context errors before introduction.
- Industry feedback focuses on policy choices, not avoidable drafting gaps.
- There's a clear audit trail showing how language was produced and approved.
What to watch next
Nally says the whisky rules could appear on the spring agenda. The team expects full reviews at each stage and is open to deciding it was a useful experiment-or something to integrate for similar files.
Either way, legislative writers aren't replaced. They get a faster starting point and more time for the work that matters: intent, coherence, and risk control.
Helpful resources
- Directive on Automated Decision-Making (Treasury Board of Canada Secretariat)
- Practical AI courses by job role (Complete AI Training)
Your membership also unlocks:
