Algorithmic Impact Assessment
A structured, documented process that identifies and evaluates the potential social, legal, and technical impacts of an algorithmic/AI system before and during its use.
How the laws define this
Regulatory frameworks agree that an algorithmic impact assessment is a structured, documented evaluation designed to identify potential risks, harms, and rights impacts across an AI system's lifecycle, establishing mitigation measures to ensure appropriate governance and oversight. The key divergence lies in legal triggers, public transparency, and operational application across instruments. Under Resolução CNJ nº 615/2025, the assessment is mandatory, must be published on Sinapses, and requires periodic reassessments and external audits for high-risk systems. By contrast, the Treasury Board of Canada Secretariat Guide applies the requirement when AI may inform administrative decisions under the Directive on Automated Decision-Making, emphasizing compliance and approval steps. Meanwhile, NITI Aayog's Approach Document frames these assessments as recommended inputs to differentiate governance intensity by risk level, guide procurement decisions, and inform sectoral guidance.
Synthesised from the 3 statutory definitions below. Each one is quoted in full, with its source.
How jurisdictions differ
What is distinctive in each definition, binding law in force first. Showing 2 of the 3 instruments that define this term; all are listed under Definitions below.
| Jurisdiction | Instrument | Status | What is distinctive |
|---|---|---|---|
| Brazil | CNJ Resolution No. 615/2025 | In Force | Requires publication of the assessment (full or redacted) on Sinapses, periodic reassessments, and external audits for high-risk systems, with a focus on fundamental rights impacts in the Brazilian Judiciary. |
| Canada | Guide on the use of generative artificial intelligence | In Force | Ties the assessment to the Directive on Automated Decision-Making for AI systems informing administrative decisions in Canadian federal institutions. |
What it triggers
- In the Brazilian Judiciary under CNJ Resolution No. 615/2025, it must be conducted before and during deployment for AI solutions, published on Sinapses, with periodic reassessments and external audits for high-risk systems.source
- In Canadian federal institutions, the Guide on the use of generative AI references a documented assessment described in the Directive on Automated Decision-Making when AI may inform administrative decisions.source
Definitions (3)
A formal, documented evaluation required before and during deployment that identifies potential impacts on fundamental rights and other risks, specifies mitigation measures, and is published (in full or redacted) on Sinapses; high‑risk systems must undergo periodic reassessments and external audits tied to these assessments.
A documented assessment required (per the Directive on Automated Decision‑Making and referenced by the Guide) when an AI system may inform administrative decisions; it identifies risks, impacts, required mitigations, documentation, and approval steps to ensure compliance and appropriate oversight.
A structured assessment to identify and evaluate potential harms, rights impacts, and risks from an AI system across its lifecycle; used to differentiate governance intensity by risk level and to prioritize mitigation and oversight measures. The Approach Document recommends AI/algorithmic impact assessments for high-impact deployments and as an input to sectoral guidance and procurement decisions.
Not to be confused with
While both evaluate rights impacts before deployment, the algorithmic impact assessment under CNJ Resolution No. 615/2025 explicitly includes fundamental rights alongside other risks and mandates publication and audits.
AI risk management is a broader systematic application of policies to identify, assess, treat, and monitor risks, whereas algorithmic impact assessment is a specific structured, documented process.
Conformity assessment verifies regulatory compliance before market placement for high-risk AI, unlike the algorithmic impact assessment's focus on identifying and evaluating social, legal, and technical impacts.
Questions and answers
What publication requirements apply to algorithmic impact assessments in Brazil?
Under CNJ Resolution No. 615/2025, the assessment must be published, either in full or redacted, on Sinapses.
When is an algorithmic impact assessment needed in Canadian federal institutions?
It is described in the Directive on Automated Decision-Making, referenced in the Guide on the use of generative AI, when an AI system may inform administrative decisions.
Do high-risk AI systems face additional obligations in the Brazilian Judiciary?
Yes, they must undergo periodic reassessments and external audits tied to the algorithmic impact assessment under CNJ Resolution No. 615/2025.
Related Terms
Fundamental Rights Impact Assessment
Evaluation of AI system's potential impact on human rights before deployment....
AI risk management
Systematic application of management policies and practices to identify, assess, treat, and monitor AI-specific risks....
Conformity Assessment
Process to verify that high-risk AI systems meet regulatory requirements before market placement....