Blog

The thinking behind AI security review.

Threat models, governance evidence, and the decisions that hold up under scrutiny. Written for security architects and the committees they report to.

Newsletter

New posts in your inbox,
when they publish.

Threat modeling, governance evidence, and what AI Committees actually need — written for security architects and AI governance leads. No cadence promises.

No spam. Unsubscribe anytime.

Technical11 min

LLM Excessive Agency — Scope Permissions Down (OWASP)

Excessive agency: an LLM has more capability than its task requires, and a manipulated model can exercise the excess. Least-privilege is the fix.

Governance10 min

Re-assessment triggers — the field most dispositions skip

A disposition without re-assessment triggers never expires. Triggers keep decisions honest as AI systems evolve. How to define them.

Technical10 min

Design-time vs runtime AI security — where review belongs

Runtime tools watch production anomalies. Design-time review decides if the system should ship. Conflating them creates gaps.

Governance9 min

When to re-assess an AI vendor

AI vendor assessments expire. Model updates, new features, changed terms, incidents, and expanding use cases all trigger a fresh review.

Reference11 min

The ISO 42001 evidence checklist for security reviews

Every evidence artefact an ISO 42001 audit will request, mapped by control domain and aligned to what AI security reviews already produce.

Technical11 min

Prompt-context injection through MCP tools

MCP tools return data injected into model context. When that data contains instructions, the tool becomes an injection vector. Controls that prevent it.

Regulation11 min

EU AI Act obligations for deployers (not just providers)

Deployers -- not just providers -- carry significant EU AI Act obligations. What organisations using AI systems for their own purposes must do.

Technical11 min

Vector Database Security — RAG Pipeline Checklist

Vector databases inherit standard data-store security requirements plus RAG-specific ones. The full security checklist for production pipelines.

Governance11 min

Human-in-the-loop boundaries that actually hold

HITL is the most common agentic control and the most often specified in ways that don't hold. What a robust boundary requires — and the failure modes.

Technical10 min

Model denial of service and cost-exhaustion attacks

LLM DoS doesn't require crashing the service — just making it expensive. Cost-exhaustion attacks are under-defended in most assessed AI systems.

Governance9 min

Who runs the AI security review — roles and hand-offs

AI security reviews span architects, security engineers, governance leads, and DPOs. Map the hand-offs to avoid dropped gates.

Governance11 min

The anatomy of an AI evidence pack

The complete artefact set a governance committee needs for a defensible AI decision. What goes in, and why order and labelling matter.