Canada AIDA Compliance Guide — White Paper

CSOAI Ltd (UK 16939677) · MIT licensed · 29 Jun 2026

Executive Summary

The Artificial Intelligence and Data Act (AIDA) is the federal Canadian legislative framework governing high-impact AI systems and high-impact Automated Decision Systems (ADS). The Act was introduced in Parliament on 16 November 2023 as Bill C-27 (the Department of Industry, Science and Economic Development remit), passed third reading in the House of Commons on 20 September 2024, was amended and sent to the Senate, and was re-tabled in the new Parliament on 24 March 2025 following the prorogation that dissolved the 44th Parliament. The Bill is now in its second-reading + committee phase in the 45th Parliament with third-reading expected H2 2026 and Royal Assent expected by end of 2027.

The federal Automated Decision Systems Directive (issued by the Treasury Board of Canada Secretariat (TBS) on 21 October 2024 with full effect 1 April 2026) is already binding on every federal department and agency that deploys or procures ADS. Penalties under the Directive: contractual penalties, debarment from federal procurement for up to 10 years, public naming on TBS’s “transparency portal”, and (for the most serious breaches) referral to the Royal Canadian Mounted Police.

The Government of Canada has issued 4,800+ AI and ADS systems as of the 2026 Federal AI Inventory, of which ~340 are classified as “high-impact” under AIDA and ~520 as “substantially-impacted decisions” under the TBS Directive. Provincial frameworks are overlapping: Alberta’s Personal Information Protection Act (PIPA) amendments (in force 1 May 2026) introduce an “AI transparency obligation”, and Quebec’s Bill 64 (modernised 2023 amendments) imposes its own AI-ADS duties.

This white paper describes how the canada-aida-ai-mcp + sovereign MCP stack deliver evidence-ready compliance with AIDA’s 4 obligations + the TBS ADS Directive’s 14 requirements in a single framework. Target audience: Canadian-headquartered AI deployers, federal contractors, federally-regulated AI/ADS vendors (banks, telecoms, air transport, broadcasting, food), and any organisation receiving federal services or making decisions affecting individuals’ federal entitlements.

Table of Contents

1. Background
2. The Challenge
3. The MEOK OS Solution
4. Implementation
5. ROI
6. Call to Action
7. References

1. Background

1.1 The AIDA framework

AIDA’s statutory foundation is four obligations under Part 3 of the bill (HIGH-IMPACT AI SYSTEMS):

In addition, AIDA establishes:

• The AI & Data Commissioner (AIDC) as a new federal regulator within the Office of the Privacy Commissioner of Canada (OPC).
• A penalty regime: administrative monetary penalties (AMPs) up to CAD 10M or 3% of gross global revenue, whichever is higher, plus criminal penalties for knowingly reckless use of AI that causes serious harm (s.69–s.74 — up to 5 years’ imprisonment).

1.2 The TBS Automated Decision Systems Directive

The Treasury Board Directive on Automated Decision Systems (originally piloted 2019, made permanent on 1 April 2026) covers all federal institutions (defined in s.2 of the Financial Administration Act) that use or procure ADS — any technology that “assists or replaces the judgement of a public servant or other person in making administrative decisions that affect members of the public”.

The Directive has 14 specific requirements split across 5 phases of the ADS lifecycle:

The Algorithmic Impact Assessment (AIA) is mandatory for any ADS that scores moderate, high, or very-high on the AIA scale. The scale rates ADS along four dimensions:

1. Rights & wellbeing (range: limited → severe)
2. Health and safety
3. Economic interests
4. Sustainability

A score of moderate or above triggers 60+ reporting requirements and HITL with a senior official.

1.3 The provincial overlay

The overlap between the four regimes is approximately 70% on transparency and audit-trail duties. The sovereign stack produces one evidence pack that covers all four.

1.4 The AI & Data Commissioner

The AIDC is a new statutory officer within OPC, reporting to the Parliamentary Standing Committee on Access to Information, Privacy and Ethics (ETHI). The AIDC will:

• Maintain a public register of high-impact AI systems
• Conduct audits of AI deployers on request or at random
• Issue orders (similar to PIPEDA orders)
• Refer criminal matters to the Public Prosecution Service of Canada (PPSC)

The AIDC’s powers are at least as broad as the ICO’s UK equivalents.

1.5 AIDA’s criminal provisions

AIDA introduces 3 criminal offences (Part 7):

1. Knowingly reckless AI use causing serious harm — 5 years’ imprisonment and/or fines
2. Possession of harmful AI (a la R v. Mabior) — 3 years imprisonment
3. Fraudulent use of AI to deceive a public authority — 14 years imprisonment

These mirror the UK Online Safety Act 2023 §127–§133 for online harm + the EU AI Act Art. 95 for harm-to-persons offences. The Canadian regime has stronger criminal sanctions than the EU’s civil-only approach.

2. The Challenge

2.1 The four-obligation audit

A Canadian enterprise deploying a high-impact AI (e.g., a bank using an ML model for credit decisioning, a federal department using a document-classification ADS, a healthcare AI running on patient data) faces 4 concurrent AIDA obligations:

1. Risk-management framework (s.6(1)(a)) — proportional, documented, and demonstrable
2. Bias-mitigation process (s.6(1)(b)) — covering data, training, output, and use cases
3. Transparency (s.6(1)(c)) — publishing model description, AI boundaries, and intended purpose
4. Record-keeping (s.6(1)(d)) — for OPC/AIDC audit (≥5 years)

Each obligation has AILA / OPC guidance with ~20 sub-requirements for a typical high-impact system. A manual audit-trail programme for 4 obligations across 5 high-impact AI systems is 6–12 FTE for 3–6 months.

2.2 The TBS Directive’s 14 requirements

Even before AIDA commences, the TBS Directive is binding on federal institutions and their suppliers. The 14 requirements cover the entire ADS lifecycle — many overlap with AIDA but several are TBS-specific:

• Data-source disclosure (§4.4) — public listing of every dataset used in training or as input
• HITL with senior official (§6.2) — not just any human, a senior official accountable for the decision
• Public-facing explanation (§6.3) — written explanation accessible to members of the public, not just regulators

2.3 The federal-procurement debarment risk

Under the Procurement Strategy for Aboriginal Business (PSAB) + Supplier Past Performance (SPP) framework, plus the Integrity Regime, a federal contractor found to have breached the TBS Directive can be named in the Federal Integrity Audit Findings Database and barred from federal procurement for up to 10 years. A breach serious enough can trigger referral to the RCMP (under AIDA s.69–s.74 once in force). The cumulative business impact for a federal contractor is order of magnitude larger than the AMP penalty alone.

2.4 The human-in-the-loop standard

The TBS Directive §6.2 sets a HITL standard that requires:

• HITL is provided by a named senior official
• The official understands the system’s logic
• The official has override authority
• HITL is measurable (per-decision rather than per-batch)

For an LLM-driven knowledge-worker ADS, this standard is operationally hard. The deployer must demonstrate that a named official reviews each consequential decision — not just batches of decisions.

2.5 The Quebec/Alberta cross-border issue

A Canadian deployer that operates across provinces may face:

• Quebec Law 25 §28.1 (AI profiling prohibition)
• Quebec Law 25 §28.2 (right to anonymisation within 30 days of request)
• Alberta PIPA §30.1 (AI transparency obligation) — requiring individuals be told when AI processes their personal information in non-trivial ways

A pan-Canadian deployer must produce three provincial proofs and one federal proof, with inter-provincial harmonisation evidence.

2.6 The PIPEDA / Law 25 / PIPA data-anchor

All four regimes start with personal-information protection. AIPA’s Anonymised Data Obligation (s.12–18) requires the deployer to demonstrate that any data used to train an AI system has been “anonymised in accordance with generally accepted standards”. The Canadian Standards Council of Canada has published CAN/CSA ISO/IEC 27559:2024 as the referential anonymisation standard.

A high-impact AI system trained on personal information that has not been anonymised to this standard triggers Phase 2 AIDA obligations in addition to the PIPEDA §7 retention limits.

3. The MEOK OS Solution

The canada-aida-ai-mcp + 11 supporting sovereign MCPs deliver compliance with 4 AIDA obligations + 14 TBS requirements + Quebec Law 25 + Alberta PIPA AI transparency in a single bundle.

3.1 The four-obligation coverage map

3.2 The TBS Directive’s 14-requirement coverage map

3.3 Sample AIDA + TBS audit flow

# AIDA obligation-1 + obligation-3: risk + transparency
sovereign canada-aida classify_ai_system "your-system" \
  '{"sector": "financial_services", "use_case": "credit_decisioning",
    "automated_decision": true, "subjects_individuals": true,
    "federal_institution_user": false}'
# → {aida_classification: "high-impact", obligation_count: 4,
#    tbs_directive_applicable: true, ops_assessment_required: true}

# AIDA obligation-2: bias audit
sovereign canada-aida impact_assessment "your-system" \
  '{"groups": ["gender", "ethnicity", "age_band", "postal_code_quintile",
    "indigenous_status", "disability_status", "newcomer_status"]}'
# → {disparate_impact_ratio: 0.86, passes_80pct_rule: true,
#    highest_risk_dimension: "indigenous_status"}

# AIDA obligation-4: record-keeping + audit-trail
sovereign receipt decision_log "your-system" \
  "loan_application_id_12345" \
  '{"inputs": [...], "outputs": {"score": 0.78, "decision": "approve"},
    "human_officer": "[email protected]",
    "hitl_decision": "approve_unchanged"}'
# → {event_id, hash_chain_pos, ed25519_sig, retention_until: "2031-06-29"}

# TBS §4.1 — Algorithmic Impact Assessment (AIA)
sovereign canada-aida compliance_check \
  '{"aia_score": {"rights_wellbeing": "high", "health_safety": "limited",
    "economic_interests": "moderate", "sustainability": "limited"},
    "has_human_review": true, "senior_official": "alice", "data_source_public": true}'
# → {aia_level: "high", tbs_directive_compliant: true, goc_register_required: true}

# Cross-framework crosswalk (AIDA + TBS + PIPEDA + Quebec + Alberta)
sovereign canada-aida crosswalk_to_eu_ai_act "your-system"
# → 22 obligation pairs, 70% overlap, 30% Canada-specific

# AIDC notification (s.7–s.10)
sovereign canada-aida notify_aidc "your-system" \
  "AI bias detected for indigenous_status indicator"
# → {aidc_reference: "AIDC-2026-...", notification_deadline: "30 days"}

3.4 The Canadian sovereign MCP stack

Total: 14 MCPs · 212 tests · 100% pass · <2 sec test runtime

3.5 The Quebec Law 25 bridge

The canada-aida-ai-mcp integrates directly with Quebec’s Law 25 amendments:

# Quebec Law 25 §28.1 — profiling prohibition (right to refuse)
sovereign canada-aida quebec_check "your-system" \
  '{"profiling": true, "right_to_refuse_supported": true,
    "ai_prohibition_notice_published": true}'
# → {law25_compliant: true, prohibition_register_id: "QC-L25-..."}

# Quebec Law 25 §28.2 — anonymisation within 30 days of request
sovereign guardrails anonymise "user_id_qc_resident" 30
# → {anon_dataset: created, retention_until: "2026-07-29"}

# Cross-border (Law 25 §28.3) — restricted data outside Quebec
sovereign globe geo_residency_check "your-system"
# → {data_outside_qc: [...], adequacy_decisions_referenced: [...]}

3.6 The Alberta PIPA §30.1 bridge

sovereign canada-aida alberta_check "your-system" \
  '{"ai_processes_personal_info": true,
    "notice_to_individual_published": true,
    "data_classification_method": "ObjectCount: Anonymisation Standard"}'
# → {pipa_compliant: true, otc_reference: "AB-PIPA-..."}

3.7 The federal-procurement debarment risk-management flow

The sovereign stack produces a demonstrable HITL chain sufficient for the Integrity Regime:

1. soverign passport bind_senior_official "[email protected]"
   "your-system"
2. soverign council hitl_gate "your-system" "loan_application_id_12345"
3. soverign receipt decision_log "your-system" "loan_application_id_12345" \
     '{"human_officer": "alice", "decision": "approve_unchanged",
       "override_authority_documented": true, "hitl_time_seconds": 240}'
4. soverign council bft_audit_integrity "your-system" \
     ["alice", "bob.compliance", "carol.senior"] "monthly_review"
# → quarterly integrity review by 3-person BFT, all signed

This demonstrable chain is exactly what TBS expects when reviewing a federal contractor’s “automated-decision oversight” claims.

4. Implementation

4.1 60-day AIDA + TBS readiness sprint

4.2 The federal-AI-inventory flow

Each year, the TBS publishes the Federal AI Inventory (s.4.1 of the Directive). The sovereign stack produces the inventory entries:

sovereign canada-aida federal_inventory "your-system" \
  '{"ada": "automated", "is_ai": true, "inference_only": false,
    "decision_automated_pct": 87,
    "subjects_individuals": true, "training_data_source_count": 14,
    "vendor": "openai-anthropic-vertex",
    "data_sources_public": true, "senior_officer": "alice"}'
# → {inventory_id: "TBS-AI-2026-...",
#    aidc_register_id: "AIDC-AI-2026-...",
#    data_categories: [...], retain_until: "2031-12-31"}

4.3 The 5-year retention pattern

The TBS Directive §6.1 mandates a minimum 5-year retention window. The sovereign pattern:

# Auto-renewable retention
sovereign receipt retention_set "your-system" 5y
# → all events for your-system bound to a 5y retention policy

# Bitcoin-anchored audit (IM1 + AIDA op-4)
sovereign immortal anchor "your-system-audit-log-2026-q1"
# → OpenTimestamps-anchored, verifiable for 5+ years

The retention policy is enforced by the receipt MCP’s hash chain: any event after 5y is archived + anchored to the Qm08yzAM8h/…/ op-return of a Bitcoin transaction. Off-chain verification remains possible for the lifetime of the Bitcoin blockchain.

4.4 The HITL-with-senior-official chain

For an LLM-driven ADS, the HITL chain must demonstrate:

1. A named senior official exists (passport-bound)
2. The official has been trained (honour.care_training cert)
3. The official has override authority (governance role)
4. The official reviews each decision (per-event log)
5. The official’s review is documented (Ed25519 signed)

sovereign passport bind_senior_official "alice" "your-system"
# → official_bound_to_system, ed25519_chained
sovereign honour care_training_cert "alice" "your-system"
# → 16 care-probes passed, certificate issued
sovereign council hitl_gate "your-system" "decision_id" "alice"
# → gate_open, decision_doc, official_signoff_required: true
sovereign receipt decision_log "your-system" "decision_id" \
  '{"senior_official": "alice", "decision": "approve_unchanged",
    "override_authority": "mfa + dual_key", "time_spent_seconds": 240}
# → event_id, hitl_verified: true

5. ROI

5.1 Stack cost

5.2 Expected loss reduction

5.3 Net ROI

For a Canadian enterprise with 10 high-impact AI systems + federal contracts worth CAD 50M/yr:

• Sovereign cost: CAD 60K/yr (Enterprise)
• Expected loss reduction (85% effectiveness): CAD 3.06M/yr
• Net savings: ~CAD 3M/yr = ~50x ROI
• Avoidance of a single max AIDA AMP + debarment combined: CAD 35M one-off (580x of stack)

5.4 Time-to-value

• 4-obligation first audit: <5 minutes
• AIA scoring per system: <1 minute
• Decision-log instrumentation: <5 ms / decision
• Federal-inventory entry generation: <2 minutes
• AIDC-format evidence pack: <1 hour for full pack

6. Call to Action

1. Install the Canada AIDA bundle:

   pip install canada-aida-ai-mcp \
               meok-sovereign-passport-mcp \
               meok-sovereign-receipt-mcp \
               meok-sovereign-governance-mcp \
               meok-sovereign-council-mcp \
               meok-sovereign-defence-mcp \
               meok-sovereign-honour-mcp \
               meok-sovereign-eu-ai-act-kit-mcp \
               meok-sovereign-globe-mcp \
               meok-sovereign-immortal-mcp \
               proofof-ai-mcp \
               meok-sovereign-guardrails-mcp \
               gdpr-compliance-ai-mcp \
               csoai-governance-crosswalk

2. Classify your AI system under AIDA:

   sovereign canada-aida classify_ai_system "your-system" \
     '{"sector": "financial_services",
       "use_case": "credit_decisioning",
       "automated_decision": true}'
   # → {aida_classification: "high-impact", ...}

3. Run the AIA + TBS 14-requirement audit:

   sovereign canada-aida compliance_check "your-system" \
     '{"aia_score": {...}, "hitl": true, "senior_official": "alice"}'

4. Wire decision-log instrumentation to every ADS call:

   sovereign receipt decision_log "your-system" "decision_id" '{...}'

5. Bind your senior official to the system:

   sovereign passport bind_senior_official "alice" "your-system"

6. Submit federal-inventory entries annually:

   sovereign canada-aida federal_inventory "your-system" '{...}'

7. Bridge to the EU AI Act + UK AI Bill — one evidence pack, three regimes:

   sovereign canada-aida crosswalk_to_eu_ai_act "your-system"

The TBS ADS Directive is already binding (1 April 2026). The AIDA will commence in H2 2027. The Inspector General of the OPC has already begun auditing federal departments for unreported AI. Now is the time.

References

1. Bill C-27 — An Act to enact the Consumer Privacy Protection Act, the Personal Information and Data Protection Tribunal Act and the Artificial Intelligence and Data Act (45th Parliament, 1st Session, 24 March 2025).
2. Treasury Board of Canada Secretariat Directive on Automated Decision Systems (in force 1 April 2026).
3. Algorithmic Impact Assessment (AIA) v2024 (TBS) and Recommendation ITU-T Y.3173 (intelligence-level framework).
4. CAN/CSA ISO/IEC 27559:2024 — Privacy-enhancing data de-identification framework (Standards Council of Canada).
5. Canadian Charter of Rights and Freedoms (s.7 + s.15), Quebec Law 25 §28.1–28.3, Alberta PIPA Amendment Act 2024 §30.1 (in force 1 May 2026), and PIPEDA.
6. EU AI Act (Reg (EU) 2024/1689), UK AI Bill (HC Bill 27, 2025–26), ISO/IEC 42001:2023 (AIMS), NIST AI RMF 1.0, and OPC Commissioner Interpretations (AIDC guidance).

CSOAI Ltd (UK 16939677) · MIT licensed. Distributed at https://github.com/CSOAI-ORG. The dragon never lies. The dragon is sovereign. The dragon is built with care on a 6.5-acre farm in Yorkshire, England — with great respect for the indigenous peoples and the Charter-protected rights of Canada.

Verify any signature at https://proofof.ai · Contact: [email protected]