From the Campus Operations edition of October 18, 2026
York University's 15 October employer AI letter of understanding commits to human review of AI-assisted employment decisions, 60-day notice for HR-related AI, and consent before training on staff data
York University; CUPE 3903 (teaching assistants, contract faculty, graduate and research assistants) · Research university: academic labor relations, HR and data governance · Toronto, Ontario, Canada
- Publisher
- York University employer proposal, published by CUPE 3903
- Original publication
- Document dated 15 October 2026 ('WITHOUT PREJUDICE'), posted on the CUPE 3903 website
- Source retrieved
- 2026-10-19
- Event date
- 2026-10-15
What happened
York's employer proposal for an all-units AI letter of understanding defines AI by reference to Ontario's Enhancing Digital Security and Trust Act. It would require AI use to be transparent to materially affected employees and subject to human oversight, and bars AI from making decisions on terms and conditions of employment without human review and final approval. It requires at least 60 days' written notice to the union before first use, or material change in use, of AI that provides input to hiring, performance evaluation or discipline, naming purpose, data and decision role. It also offers training on approved tools, bars knowingly submitting employee personal information to unvetted AI, and requires express consent before employee data or IP is used to train or fine-tune models, while excluding retrieval-only processing from that consent rule. Why now: the archive holds only York's separate same-day counter-proposal for a joint working group; this document is new and shows what the employer has put in writing.
Why it matters
Gives campus HR and IT a concrete, employer-drafted template for AI terms: human final approval on employment decisions, a fixed notice period scoped to HR uses, and a precise line between model training (consent needed) and retrieval (no consent).
Evidence and measured results
Primary text of a without-prejudice bargaining proposal. It is not agreed language, and the union's response is not included. The relationship between this proposal and the same-day joint-working-group counter-proposal is not explained in the document.
Limitations and uncertainty
Proposal only; terms may change or be dropped. Applies to CUPE 3903 units, not all York staff. Read alongside the archived York counter-proposal; this record does not establish which text will prevail.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-18; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Universities need AI terms they can live with at the bargaining table and then actually operate.
- Stakeholders
- VP people and culture
- Labor relations
- CIO and privacy officer
- General counsel
- Union bargaining teams
- Discovery questions
- Which AI systems give input to hiring, evaluation or discipline today?
- Could you meet a 60-day notice window for HR-related AI?
- Do your vendors train on staff data, and can you show they do not?
- Value hypothesis
- Interpretation: an HR-AI register and notice procedure prepared before agreement could shorten implementation once terms settle.
- Potential engagement
- An HR-AI system register, notice template and vendor training-use review for one institution.
- Unsupported claims
- That the union accepts these terms
- Any effect on grievances or costs
- That the counter-proposal has been withdrawn
Pre-sales engineering
Role takeaway
- Fit
- Relevant to AI in HR, recruitment, performance and any assistant processing staff data.
- Architecture and integration
- Separate retrieval-only processing from model training on staff data; log human approval in HR workflows.
- Prerequisites
- Vendor documentation of training use
- Privacy and security vetting process
- HR workflow audit logs
- Constraints
- Ontario statutory AI definition
- Proposal language may change
- Consent required before training on staff data
- Security
- Personal information only to vetted systems; express consent before training or fine-tuning on employee data or IP.
- Proof of value
- Collect vendor attestations on training use for each HR-related tool
- Demonstrate an audit record of human final approval
Delivery
Role takeaway
- Work
- Build a 60-day notice step and decision-role description into HR technology release plans.
- Dependencies
- Ratified agreement
- Vendor transparency
- HR and IT coordination
- Ownership
- HR and labor relations own notice; IT and privacy own vetting.
- Skills and adoption
- Training on approved AI tools for affected staff, as the proposal offers.
- Governance checkpoints
- Privacy vetting before staff data is used
- Union notice 60 days before go-live
- Human approval recorded for employment decisions
- Acceptance criteria
- Proposed: 100% of HR-related AI noticed 60 days ahead
- Proposed: no model training on staff data without recorded consent
- Risks
- Terms change before agreement
- Vendors cannot evidence training use
- Notice window slows urgent fixes
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
The training-versus-retrieval distinction maps onto real architectures: retrieval-augmented assistants over HR data would need vetting but not consent, while fine-tuning on staff content would. Systems must be able to show which they do.
Governance
Who approves, reviews and stays accountable for outcomes?
Notice applies only to AI used for hiring, evaluation or discipline, narrower than union demands seen elsewhere; human review attaches to employment decisions.
Security and privacy
What data, permissions and controls need testing?
Employee personal information may only go to AI systems vetted under university privacy and security policies; training on employee data or metadata requires express informed consent.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Offers training on approved AI tools and human review of AI-assisted employment decisions; it does not address job displacement, which other unions are seeking.
Procurement
What should contracts, pricing and exit terms secure?
Vendors would need to state whether customer data trains models and support a 60-day notice window before HR-related AI goes live.
Operating model
Which teams own the service once it runs?
HR technology changes would gain a fixed lead time and a documented decision-role description for each system.
Publication history
- 2026-10-18Campus Operations · Issue 417 resources
Stable resource ID: york-employer-proposal-ai-workplace-lou-human-review-notice-20261015