Notifications
Clear all
Topic starter
18/08/2026 2:36 pm
TL;DR: AI model security has become a lifecycle problem, not just an infrastructure problem, because models, training data, prompts, APIs, and connected identities all expand the attack surface, according to BigID. The governance gap is that traditional controls do not fully address probabilistic behaviour, data leakage, and non-human access across AI systems.
NHIMG editorial — based on content published by BigID: AI model security and governance across the AI lifecycle
By the numbers:
- 88% of organizations now use AI in at least one business function, up from 78% the previous year.
- AI agents and non-human identities are part of the security surface, and 80% of identity breaches involved compromised non-human identities such as service accounts and API keys.
- Only 5.7% of organisations have full visibility into their service accounts.
Questions worth separating out
Q: What problem does ownership attribution solve for service accounts and API keys?
A: It closes the gap between exposure detection and accountable remediation.
Q: Why do AI systems create NHI governance problems?
A: AI systems often rely on service accounts, tokens, APIs, and delegated permissions that behave like non-human identities.
Q: What is the difference between prompt injection and model theft?
A: Prompt injection changes what an AI system does by steering its runtime behaviour, while model theft tries to reconstruct the model’s capabilities through repeated queries or probing.
Practitioner guidance
- Inventory AI-connected identities Catalog every service account, API key, token, and agent that can reach models, prompts, training data, or vector stores.
- Separate controls by lifecycle stage Apply different controls for training data, model artefacts, deployment, and runtime inference.
- Constrain model and agent privilege Reduce the permissions granted to AI systems and the identities behind them.
What's in the full article
BigID's full article covers the operational detail this post intentionally leaves for the source:
- The article walks through model, data, prompt, and deployment risks in more depth, including how each threat manifests across the AI lifecycle.
- It includes practical examples of data poisoning, prompt injection, model theft, and access abuse that teams can use in internal threat modelling.
- The source explains how BigID positions discovery, classification, and governance across models, agents, prompts, and pipelines.
- It expands on the framework mapping to NIST AI RMF, OWASP GenAI, and Google Secure AI Framework guidance.
👉 Read BigID's analysis of AI model security across the full AI lifecycle →
AI model security and identity controls: what teams need to know?
Explore further
18/08/2026 3:21 pm
AI model security is now an identity problem as much as a model problem. The article correctly shows that models, prompts, APIs, and data are only secure when the identities that reach them are governed. That means service accounts, API keys, and AI agents need the same lifecycle discipline applied to other high-risk credentials. Practitioners should treat AI access as part of IAM and PAM, not as a separate AI-only control plane.
A question worth separating out:
Q: How can organisations tell whether their AI security model is actually working?
A: They should test whether the control stack can explain who acted, what data was touched, and what purpose the action served. If those three signals cannot be correlated in one incident view, the model is likely monitoring access without governing behaviour. That is a visibility gap, not a complete AI security posture.
👉 Read our full editorial: AI model security now depends on identity, data, and lifecycle controls
