CoreQ Academy1.2 Explain the importance of data security in relation to AI.
An AI system is only as trustworthy as its data. Bad, biased or tampered data produces bad decisions, and data fed into AI can leak back out.
Data processing
- Data cleansing
- Fixing or removing errors, duplicates and junk before training. Also an opportunity to strip secrets and personal data.
- Data verification
- Checking that data is accurate, complete and in the expected format before use.
- Data lineage
- A record of where data came from and every transformation it went through. Lets you trace a bad model output back to bad input.
- Data integrity
- Assurance that data has not been altered without authorisation, using hashes, signatures and access control.
- Data provenance
- Proof of the origin and ownership of data, including licensing and consent. Critical for legal use and for spotting untrusted sources.
- Data augmentation
- Creating additional training examples by transforming existing ones (rotating images, paraphrasing text) to improve robustness.
- Data balancing
- Adjusting the mix of classes so the model is not biased toward the majority, e.g. far more benign than malicious samples.
Data types
- Structured data
- Organised in fixed fields and rows: databases, spreadsheets, log tables.
- Semi-structured data
- Has tags or keys but no rigid schema: JSON, XML, email headers.
- Unstructured data
- No predefined format: documents, images, chat transcripts. Hardest to classify and scan for sensitive content.
Watermarking
- Watermarking
- Embedding a detectable signal in AI-generated content (or in a model) to identify its origin. Helps detect synthetic media and prove model ownership; can sometimes be removed.
Retrieval-augmented generation (RAG)
- Retrieval-augmented generation (RAG)
- The model retrieves relevant documents from a knowledge store at query time and uses them to answer. Keeps answers current without retraining, but the retrieved content becomes an attack path (indirect prompt injection) and must respect the user's permissions.
- Vector storage
- A database that stores embeddings and finds the most similar ones quickly. Needs access control, encryption and per-user filtering like any other data store.
- Embeddings
- Numeric vectors that capture the meaning of text or images so similar items are close together. Embeddings can leak information about the source text, so treat them as sensitive.
An internal RAG assistant answers an intern's question by quoting the CFO's salary spreadsheet. What went wrong?
Show answer
Retrieval ignored the user's permissions. The vector store must enforce document-level access control (filter by the requester's entitlements) before content reaches the model.
Check your understanding
Loading quiz…
Original CoreQ Academy material, mapped to objective 1.2 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.