Field notes

Notes from the AI security frontline.

Research, deployment guidance and control design from the team building XAIGuard. No hype, no invented benchmarks — just what works when you have to secure AI in production.

4 articles · RSS

Filter

ComplianceEvidenceAudit

What audit-grade AI evidence actually looks like

Hash-linked entries, Ed25519-signed exports, scoped auditor access and verifiable chains — and the structural reason a folder of screenshots will never pass.

06 Aug 2026 · 9 min read

AgentsMCPGovernance

Governing autonomous agents: drift, blast radius and containment

An agent is a program with a budget, a toolset and a tendency to change. Governing it means baselining behaviour, bounding what it can reach, and containing it automatically when it steps outside.

16 Jul 2026 · 10 min read

DeploymentDLPRuntime

Prompt DLP without a firewall change: three deployment paths compared

PAC file, browser extension, or API proxy. Each covers a different slice of AI traffic, and the order you roll them out matters more than which one you start with.

25 Jun 2026 · 10 min read

Shadow AIDiscoveryData Posture

Shadow AI is already inside your perimeter — here's how to find it

Proxy, firewall and CASB logs already contain a record of every GenAI destination your people reach. Here is how to turn those logs into an inventory, and what to do with it once you have one.

11 Jun 2026 · 9 min read