Databricks Certified Machine Learning Associate
Security Governance and Responsible AI
Apply security, privacy, compliance, and responsible AI controls to exam scenarios.
Official Scope and Verification
This lesson is mapped to the verified Databricks Certified Machine Learning Associate outline. Official sources and public status were rechecked on 2026-07-13. Provider pages remain authoritative for late-breaking blueprint, availability, scheduling, price, language, delivery, and retake changes.
Current Databricks proctored certification with published domain percentages.
Official Objectives Emphasized Here
| Domain or objective area | Published weight | Key objective groups | Official source |
|---|---|---|---|
| Databricks Machine Learning | 38% | Identify the best practices of an MLOps strategy; Identify the advantages of using ML runtimes; Identify how AutoML facilitates model and feature selection; Identify the advantages AutoML brings to the model development process; Identify benefits of account-level Unity Catalog feature store tables over workspace-level tables; Create a feature store table in Unity Catalog; Write data to a feature store table; Train a model with features from a feature store table; Score a model using features from a feature store table; Describe the differences between online and offline feature tables; Identify the best run using the MLflow Client API; Manually log metrics, artifacts, and models in an MLflow Run; Identify information available in the MLflow UI; Register a model using the MLflow Client API in the Unity Catalog registry; Identify benefits of registering models in the Unity Catalog registry over the workspace registry; Identify when promoting code is preferred over promoting models and vice versa; Set or remove a tag for a model; Promote a challenger model to a champion model using aliases | Databricks official Machine Learning Associate exam guide PDF |
| Data Processing | 19% | Compute summary statistics on a Spark DataFrame using .summary() or dbutils data summaries; Remove outliers from a Spark DataFrame based on standard deviation or IQR; Create visualizations for categorical or continuous features; Compare two categorical or two continuous features using the appropriate method; Compare and contrast imputing missing values with the mean, median, or mode; Impute missing values with the mode, mean, or median value; Use one-hot encoding for categorical features; Identify model types or data sets where one-hot encoding is or is not appropriate; Identify scenarios where log scale transformation is appropriate | Databricks official Machine Learning Associate exam guide PDF |
| Model Development | 31% | Use ML foundations to select the appropriate algorithm for a given model scenario; Identify methods to mitigate data imbalance in training data; Compare estimators and transformers; Develop a training pipeline; Use Hyperopt fmin to tune model hyperparameters; Perform random, grid, or Bayesian search for hyperparameter tuning; Parallelize single node models for hyperparameter tuning; Describe benefits and downsides of cross-validation over a train-validation split; Perform cross-validation as part of model fitting; Identify the number of models trained in grid-search and cross-validation; Use common classification metrics such as F1, log loss, and ROC/AUC; Use common regression metrics such as RMSE, MAE, and R-squared; Choose the most appropriate metric for a scenario objective; Exponentiate log-transformed variables before calculating metrics or interpreting predictions; Assess the impact of model complexity and the bias-variance tradeoff on model performance | Databricks official Machine Learning Associate exam guide PDF |
| Model Deployment | 12% | Identify differences and advantages of batch, realtime, and streaming model serving approaches; Deploy a custom model to a model endpoint; Use pandas to perform batch inference; Identify how streaming inference is performed with Delta Live Tables; Deploy and query a model for realtime inference; Split data between endpoints for realtime inference | Databricks official Machine Learning Associate exam guide PDF |
Authoritative Sources for This Scope
- Databricks official Machine Learning Associate exam guide PDF - Official source; accessed 2026-07-13.
Security, governance, and responsible AI questions ask whether the solution can be trusted, controlled, and explained. For Databricks Certified Machine Learning Associate, treat governance as part of the design, not a separate cleanup task after the model works.
Controls To Recognize
| Control area | What it protects | What to look for in a scenario |
|---|---|---|
| Identity and access | Systems, documents, tools, models, and administrative actions. | Least privilege, role-based access, service identities, approval boundaries, and separation of duties. |
| Data protection | Training data, prompts, uploaded files, retrieved documents, logs, and outputs. | Classification, encryption, masking, retention, residency, and deletion requirements. |
| Output quality and safety | Users, customers, business decisions, and public trust. | Grounding, citations, evaluations, content filters, policy checks, and human review. |
| Responsible AI | Fairness, transparency, accountability, and social impact. | Bias testing, explainability, consent, documentation, stakeholder review, and appeal paths. |
| Auditability | Evidence that the system was governed and operated responsibly. | Logs, versioning, approvals, risk registers, control tests, and incident records. |
Provider-Specific Risk Lens
Govern tables, features, vectors, models, notebooks, endpoints, service principals, lineage, and workspace access.
For Databricks, a governance answer is strongest when it matches the provider's identity model, logging approach, data controls, and official responsible AI guidance instead of describing safety in general terms only.
Track-Specific Risk Checks
- privacy leakage through prompts, files, logs, retrieved documents, or generated outputs
- hallucinated or ungrounded answers used without review
- unclear accountability when an AI recommendation affects people, money, security, or compliance
- training-serving skew
- data leakage between train and test sets
- model drift and stale features
Responsible AI Scenario Checklist
- Purpose: Is the use case appropriate, useful, and clearly bounded?
- People: Who is affected, who can challenge the output, and who owns the decision?
- Data: Was the data collected, used, stored, and shared appropriately?
- Model behavior: Are hallucination, bias, toxicity, privacy leakage, and misuse tested?
- Operations: Are monitoring, incident response, change control, and retirement plans defined?
Example: Prompt Injection And Data Leakage
Scenario: an AI assistant can read internal knowledge articles and call workflow tools. A user tries to make it ignore its instructions and reveal restricted information. The best answer is not just 'write a better prompt.' It should combine access control, tool permission limits, input and output filtering, retrieval permissions, logging, testing, and human escalation for sensitive actions.
How To Study Governance
- Write one governance control for each lifecycle stage: design, data, build, test, deploy, monitor, and retire.
- Practice rejecting answers that rely on user trust, prompt wording, or policy documents without enforcement.
- Use NIST AI RMF and OWASP GenAI security resources as general reference points, then map them back to the provider-specific credential objectives.
Useful Links
- Databricks Certification and Badging - Official Databricks certification and accreditation catalog.
- Databricks Academy - Official learning platform entry point.
- NIST AI Risk Management Framework - General reference for AI risk management practices.
- OWASP GenAI Security Project - General reference for LLM and GenAI application risks.