Security engineering you can read before you buy — research, tooling, and field guides. No gated PDFs, no lead-capture walls.
The working method behind a CloudSecOps AWS security assessment: 69 checks organised by attack path rather than by service, the evidence request, read-only collection, a severity rubric, and why a benchmark-compliant account can still be three hops from account-wide admin.
· 113 min read
A repeatable methodology for testing prompt injection in LLM, RAG and agent systems: rules of engagement, test matrix, trial protocols under non-determinism, severity, evidence capture and retest.
· 73 min read
A threat model for Model Context Protocol deployments pinned to specification revision 2026-07-28: trust boundaries, a 41-row threat register with attacker capability and impact per row, three attack trees, and an account of what MCP declines to defend.
· 75 min read
Writing the constraint is the easy part. An operational guide to running Gatekeeper as admission control you can afford to have fail: fail-open defaults, cold caches, audit blind spots, staged rollout, and the silent-drop bugs that make a green CI run mean nothing.
· 63 min read
Thirty-six field notes on DevSecOps practice: pipeline identity, build provenance, secret blast radius, and vulnerability triage after NVD stopped enriching a third of new CVEs. Each note states the condition under which it stops being true.
· 65 min read
Most evidence you need to detect attacks on AI workloads is off by default, billable, capped or sampled. A field guide to the telemetry that exists today.
· 77 min read
An engineering guide to automated security remediation on AWS: when to automate, why an organization trail plus a service control policy removes most of the CloudTrail-disruption problem before any code runs, and how to build a loop that is idempotent, bounded, reversible and switchable-off.
· 55 min read
Translate AI governance into systems you can build: inventory reconciled against runtime telemetry, deployment gates that block, human oversight you can measure, and evidence that survives review. With EU AI Act dates as amended in July 2026.
· 77 min read
8 articles · RSS