CoreQ Academy2.3 Given a scenario, implement appropriate access controls for AI systems.
AI systems add new identities and new paths to data. Apply least privilege to who can use a model, what data it can see, what an agent may do and which networks can reach it.
Access control areas
- Model access
- Who can query, fine-tune, download or replace a model. Model weights are valuable IP; restrict and log access.
- Data access
- The model, RAG pipeline and users should only reach data the requesting user is entitled to. Enforce permissions at retrieval, not in the prompt.
- Agent access
- AI agents that call tools act with real permissions. Give each agent its own scoped identity, least-privilege tools, short-lived credentials and human approval for high-impact actions.
- Network/application programming interface (API) access
- Restrict which networks and clients can reach model endpoints: private endpoints, API keys or OAuth, mutual TLS, IP allow lists, and keys stored in a secrets manager.
An AI agent that summarises tickets has a service account with admin rights to the whole ticketing system and can delete tickets. What should change?
Show answer
Give the agent its own identity with read-only access to the ticket fields it needs, no delete permission, short-lived credentials, and require human approval for any write action.
Check your understanding
Loading quiz…
Original CoreQ Academy material, mapped to objective 2.3 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.