CoreQ Academy2.4 Given a scenario, implement data security controls for AI systems.
Data flows into AI through training, prompts and retrieval, and flows out through responses and logs. Protect it at every step and send the model as little sensitive data as possible.
Encryption requirements
- In transit
- TLS between users, gateways, models, vector stores and tools.
- At rest
- Encrypt training data, vector databases, model weights, logs and backups with managed keys.
- In use
- Protect data while processed using confidential computing or secure enclaves where the risk justifies it.
Data safety
- Data anonymization
- Removing or transforming identifiers so data cannot be linked back to a person. Beware re-identification from combined attributes.
- Data classification labels
- Tagging data by sensitivity so pipelines and gateways can decide what may be sent to which model (e.g. no Restricted data to public models).
- Data redaction
- Removing sensitive values (names, card numbers, secrets) from prompts, documents or logs before they reach the model or storage.
- Data masking
- Replacing sensitive values with realistic placeholders, e.g. for test datasets or display.
- Data minimization
- Collect, send and retain only the data actually needed for the task.
Check your understanding
Loading quiz…
Original CoreQ Academy material, mapped to objective 2.4 of the publicly available CompTIA SecAI+ CY0-001 exam objectives (Exam objectives V1, document version 1.1). Not affiliated with or endorsed by CompTIA. Contains no exam questions.