ππ¨π¬π ππ π¬π²π¬πππ¦π¬ ππ«π π§π¨π βπ‘π’π π‘-π«π’π¬π€β ππ² πππ¬π’π π§.
Donβt be generic in risk assessment, rather the micro level dynamics of stack of an AI System needs to be focussed upon to have actual perspective of risk assessment.
Hence, if not taken care of, they become high-risk by architecture.
Traditional AI, Agentic AI, and Agentic RAG are being regulated very differently because control, memory, autonomy, and evidence now matter more than models.
This is where AI governance has quietly changed.
EU AI Act risk tiers
Unacceptable| High-Risk| Limited-RiskΒ
(Prohibited) Risk depends on use-case and domain (e.g. employment, credit, biometrics)
India Governance: DPDPA + sectoral laws + government AI advisoriesIndia: Use-case impact classification (Internal)
NIST AI RMF, ISO/IEC 42001
Compute-related AI advisors
IndiaAI Compute
Sectoral laws RBI, SEBI, IRDAI
TRADITIONAL AI(Fixed Pipeline)
Fixed Pipeline High-Risk
Specify Task
Collect Data
Refine Data
Feature Store (offline)
Model Training (offline)
Deploy Model -> Evaluate Outcomes
Typical controls:
dataset governance, model testing, security + documentation, (model cards), audit logs
UPSICI
AGENTIC AI(Compounding Learning – Bridge)
Set Objectives Limited-Risk / High-Risk
Choose LLM Model
Integrate Tools & APIs
Embed Logic + Iterations
Reason – Plan – Act – Observe
Improve
Typical controls: IISG 42001
Tool-use controls, human oversight,
action logging, approvals, incident response
NIST AI RMF, ISO/IEC 42001
AGENTIC RAG(Memory + Retrieval – Individual & Collective)
India : Use-case impact classification Set Task Goal Limited-Risk / High-Risk
Fetch Useful Data (Vector Search/API)
Query Decomposition
Apply Iterative Logic (Apply RAG)
Write to Long-Term Memory (optional) (Query ACS)
Produce + Verify Results Apt for Future Use
Typical controls:
Data access governance, task goal logging, oversight, data minimization, retention/ deletion, maried logs, audit logs.
DPDPA 2023 + DPDP Rules 2025, GDPR
IndiaAI Mission, IT Act, IT Rules
This blog is an academic initiative brought to you by the Data Privacy Pro team of AMLEGALS. Subscribe β Stay updated, Stay compliant.