As of September 2026, Microsoft has signed a new AI Safety and Privacy Standard (the Standard) that establishes detailed contractual protections for certain AI products used in U.S. schools.
The Standard does not presently give Canadian school boards a right to those protections, but it indicates what Microsoft has accepted as contractually feasible for educational AI. American school boards can use the Standard to:
- seek equivalent protections from Microsoft and other AI vendors;
- strengthen conventional data protection agreements for AI services; and
- inform their own AI governance, particularly for student-facing and higher-risk uses.
Canadian school boards should pay particular attention to the Standard’s provisions on AI training, strict data minimization, profiling, AI memory and personalization, human oversight, AI companions, agentic AI and material product changes.
What is the standard?
The Standard is contained in a binding memorandum of agreement between the National Academy for AI Instruction and Microsoft Corporation. Microsoft is currently the only participating AI provider identified in the signed agreement.
Beginning Nov. 1, 2026, U.S. school boards may ask Microsoft to incorporate protections meeting or exceeding the Standard into their existing Microsoft agreements. A school board does not become a party to the Academy’s agreement: its rights arise under its own amended agreement with Microsoft.
The Standard applies to authenticated generative AI products primarily designed and marketed for education, and excludes general-purpose productivity, collaboration, search, cloud and workplace-assistance products.
The Standard contains ten principles addressing:
- use of student and educator data for AI training;
- data minimization;
- customer control, retention and deletion;
- human oversight and explainability;
- accountability and enforcement;
- security;
- transparency for families;
- equity and accessibility;
- feature changes and vendor lock-in; and
- continuing responsibility for data after termination.
No training on school data
The Standard prohibits using student, educator or customer data to develop, fine-tune, update, benchmark or otherwise improve an AI model. The prohibition cannot be avoided by de-identifying or aggregating the data. School data also cannot be used to create synthetic training data. The restriction applies to historical data, extends to subprocessors and survives termination indefinitely.
Microsoft must provide, on request, an annual confirmation signed by a senior officer that school data was not used for AI training outside a narrow safety exception, and that such data or AI learnings did not cross from the exempted safety systems into other models or products.
Strict data minimization
The Standard permits the vendor to process only data “strictly necessary” to provide the specific educational product.
The Standard also restricts behavioural tracking, keystroke logging, passive attention monitoring, long-term profiling, precise location tracking, biometrics, gaze tracking and emotional inference.
Data minimization is integral to the safety of AI products. Data from students and educators not only assists AI to learn, but is also one mechanism of understanding human potential, and in the case of a school board, the human potential of students residing in that district.
Control over changing products
The Standard recognizes that AI products evolve during the contract term. Material new features involving new data, new processing or autonomous action require notice, and the vendor must allow the board’s administrators to enable or disable them.
Significance for Canadian school boards
Canadian availability and product scope remain unclear
While Microsoft’s express commitment applies only to U.S. educational entities, Canadian school boards can reasonably ask Microsoft and other AI vendors to explain why equivalent protections should not apply to Canadian students and educators.
Knowing that the Standard excludes AI products not primarily designed for education, Microsoft may argue that ordinary Microsoft 365 Copilot and Copilot Chat are outside the Standard’s scope. School boards should require clarification over the products and features covered by protections equivalent to those found in the Standard when contracting with Microsoft.
The Standard is valuable contractual leverage
Microsoft has accepted that detailed restrictions on training, derived data, memory, profiling, product changes and agentic functionality can be addressed by contract.
School boards should give the provision forbidding the use of data to train AI special attention. The Standard reveals that a comprehensive AI training clause might address:
- fine-tuning and benchmarking;
- prompts, outputs and metadata;
- de-identified, aggregate and derived data;
- synthetic data generation;
- product improvement; and
- subprocessors.
The Standard may influence expectations
While the Standard does not establish Canadian law, it does identify emerging areas of risk that school boards should understand for the purpose of policy development and potential liability.
For example, principle 4 states that educators are prohibited from using AI for:
- automated grading without human review;
- automated disciplinary decisions or recommendations;
- automated placement, ability grouping or special education referrals;
- emotional or psychological assessment; and
- surveillance for behavioural prediction.
The Standard also:
- prohibits the use of AI companions designed to foster emotional attachment or dependency, simulate friendship, prolong engagement beyond the learning task, or encourage sensitive disclosure;
- requires agentic AI features to be capable of being turned off for students; where those features can take action outside the provider’s systems, they must be off by default unless an authorized administrator enables them with limited permissions, human oversight and audit logging; and
- requires the customer to disable persistent memory or personalization that is not governed by the customer’s configured retention and deletion policies or equivalent controls. Where the product retains student data between sessions, it requires individuals, upon request and where applicable, to be able to see and erase the stored information, and turn off memory without losing access to the service.
These provisions may inform what parents, teaching staff, school principals, trustees, auditors and regulators expect school boards to have considered, even where the Standard itself does not apply.
Conclusion
AI contracting, governance, operational implementation and oversight should be addressed together. Strong contractual terms will not correct inappropriate application or weak oversight, and sound governance parameters will not compensate for inadequate vendor commitments and implementation. A robust governance framework and compliant contractual terms will still require ongoing oversight to ensure safety in a changing regulatory and liability landscape.
BLG lawyers would be pleased to assist your school boards with AI governance, privacy and security assessments, Microsoft Copilot deployments and AI-specific vendor contracting. Reach out to the authors or any of the key contacts below.
By Borden Ladner Gervais LLP “BLG” >>
“As Canada’s law firm, BLG provides high-value advice and advocacy to address our clients’ business challenges and problems. We go beyond legal to anticipate, consult and advise in a rapidly changing digital world.
We have extensive experience acting in specialized and complex deals and disputes. Vigilant, curious and collaborative, we harness technology and innovation to offer our clients exceptional service and value.
With 800+ lawyers across Canada, we serve clients throughout North America, Europe, and Asia. Offering expertise in intellectual property, disputes and corporate transactional matters, our connectivity gives our clients the next-level service required to achieve success in complex and international matters.”
Please visit the firm link to site