Resource Center
Resource Center

Security in the context of AI

Everything you need to know about security in the context of ai. Browse curated guides, step-by-step walkthroughs, and reference material.

Smarketing Security

Security in the Context of AI

SmaIA will use AI without giving the AI provider direct access to Smarketing data or business systems. Smarketing will remain the control point for identity, permissions, data selection, actions, audit and provider configuration. This separation is the core security decision: the model can help interpret a request, but it cannot decide what a user is allowed to see or change.

Tap a section to expand

The objective is to make AI useful across Smarketing while preserving the same company separation and product permissions that protect ordinary application access. For every AI-assisted request, the system must answer five questions before data is accessed or changed:

  1. 1Who is the authenticated user?
  2. 2Which Smarketing organisation does the user belong to?
  3. 3Which product area, record and operation is the user allowed to access?
  4. 4What is the minimum information required by the AI provider?
  5. 5Is the request read-only, or does it require an explicitly approved change?

The AI model is outside the authorisation boundary. Its output is treated as a proposal that must pass deterministic application controls.

References

  1. [1]D. L. Parnas, On the Criteria To Be Used in Decomposing Systems into Modules - Foundational paper on modularity and information hiding.
  2. [2]J. H. Saltzer and M. D. Schroeder, The Protection of Information in Computer Systems - Foundational security principles including economy of mechanism, complete mediation, separation of privilege and least privilege.
  3. [3]NIST SP 800-207, Zero Trust Architecture - Identity- and policy-based access decisions without implicit trust.
  4. [4]OWASP Threat Modeling Cheat Sheet - Guidance on data flows, trust boundaries, data stores and external entities.
  5. [5]OWASP LLM01, Prompt Injection - Risks and mitigations for direct and indirect prompt injection.
  6. [6]OWASP LLM02, Sensitive Information Disclosure - Controls for reducing sensitive information exposure.
  7. [7]OWASP LLM06, Excessive Agency - Minimum tools, minimum permissions and minimum autonomy for AI systems.
  8. [8]OWASP LLM10, Unbounded Consumption - Resource and usage controls for AI systems.
  9. [9]NIST SP 800-18 Rev. 2, Developing Security, Privacy, and Supply Chain Risk Management Plans for Systems - Structure for documenting boundaries, requirements, connections and controls.
  10. [10]NIST SP 800-218, Secure Software Development Framework - Secure development, design decisions and verification practices.
  11. [11]NIST AI 600-1, Generative Artificial Intelligence Profile - Risk management guidance for generative AI systems and third-party providers.
  12. [12]How Dataflow Diagrams Impact Software Security Analysis: an Empirical Experiment - Experimental evidence that data-flow diagrams can improve security-analysis correctness.