Notifications
Clear all
Topic starter
12/06/2026 12:39 am
TL;DR: Securing LLM usage now spans API key management, request-level policy enforcement, prompt and response monitoring, and logging because risks appear at every stage of the lifecycle, according to Lasso Security. The governance gap is no longer secret protection alone but end-to-end control over how model access is granted, used, and observed.
NHIMG editorial — based on content published by Lasso Security: How to secure your entire LLM lifecycle
Questions worth separating out
Q: How should security teams govern access to production LLMs?
A: Security teams should govern production LLM access the same way they govern other sensitive non-human identities: scope credentials tightly, bind them to a specific workload or team, and enforce policy at request time.
Q: Why do API keys alone fail to secure LLM applications?
A: API keys fail when they are treated as the whole control plane.
Q: What do security teams get wrong about prompt injection risk?
A: Teams often treat prompt injection as a content-filtering problem, but it is really a trust-boundary problem.
Practitioner guidance
- Scope every model credential to a bounded use case Issue virtual keys by team, environment, or application function, then restrict each credential to the minimum provider and model set required.
- Enforce request-level policy before the model is reached Evaluate user, application, and metadata signals at the gateway so policy decisions happen before the prompt enters the LLM.
- Add behavioural detection for prompt and response anomalies Inspect prompts and outputs for jailbreak patterns, crafted injections, and sensitive-data leakage, then route high-risk events to review or blocking.
What's in the full article
Lasso Security's full blog post covers the operational detail this post intentionally leaves for the source:
- The gateway-side mechanics for issuing virtual keys by team, environment, or use case
- The specific guardrail and policy decision flow used before prompts reach the model
- The detailed observability fields captured for each verdict, including timing and check outcomes
- The combined workflow for blocking, flagging, and reviewing risky interactions
👉 Read Lasso Security's guidance on securing the full LLM lifecycle →
LLM lifecycle security: where IAM controls need to go further?
Explore further
12/06/2026 10:02 am
LLM security is becoming an identity governance problem, not just an application hardening problem. Once a model is used in production, access is no longer a single secret check. It becomes a chain of entitlements, request context, content policy, and logging discipline that must all hold together. The practical conclusion is that IAM and NHI teams need to treat model calls as governed access events, not just API traffic.
A few things that frame the scale:
- 64% of valid secrets leaked in 2022 are still valid and exploitable today, according to The State of Secrets Sprawl 2026.
- AI-related credential leaks surged 81.5% year-over-year in 2025, with the surrounding AI infrastructure leaking 5x faster than core LLM providers, according to The State of Secrets Sprawl 2026.
A question worth separating out:
Q: Should organisations log every LLM request and response?
A: Yes, but only if the logs are explainable and actionable. Security teams need to know which policy checks ran, which ones passed or failed, and what data influenced the verdict. Without that evidence, governance cannot prove control effectiveness or distinguish real attacks from harmless use.
👉 Read our full editorial: Securing the full LLM lifecycle requires layered identity controls
