AI Compliance & Cyber Brief

ISO 42001 certification is showing up in roughly 40% of enterprise AI vendor RFPs in the EU and about 25% in North America — AWS, Anthropic, and Microsoft already hold it. Vendors without it are losing deals on paperwork, not technical merit.

The 5-question ISO 42001 gut check: walk a board through what the standard actually requires versus what most companies assume their existing AI policy already covers.

If I asked your top three AI vendors for their ISO 42001 certificate today, how many could produce one — and do you know what you’d do if none of them could?

NIST released a concept note in April 2026 for an AI RMF Profile on Trustworthy AI in Critical Infrastructure (energy, water, healthcare, financial services). It’s early-stage, but it signals where the standard of care is headed for every sector — profiles, not just principles.

NIST AI RMF isn’t finished — here’s what the next chapter looks like, and why your industry’s profile is probably next.

Of the four NIST AI RMF functions — govern, map, measure, manage — which one does your organization actually have a named owner for today?

Mature programs now map red-teaming against three overlapping frameworks — NIST AI RMF, OWASP’s LLM Top 10, and MITRE ATLAS — and organizations with mature AI security testing report roughly $1.9M lower cost per incident. Testing has demonstrable ROI now, not just a compliance checkbox.

Your AI pilot needs a red team before it needs a rollout plan — most pilots skip adversarial testing entirely.

Has anyone actually tried to break your AI system on purpose — or is “testing” still just checking that it gives the right answer?

98% of organizations report employees using unsanctioned AI tools, yet only 37% have any policy to even detect it. Gartner expects task-specific AI agents in 40% of enterprise applications by year-end, up from under 5% a year ago. Visibility isn’t keeping pace with adoption.

The AI tool your team is already using without telling you — how employees quietly adopt technology under the radar.

If I asked three random employees what AI tools they used this week that IT doesn’t know about, what would they tell me?

OWASP still ranks prompt injection as the top risk to AI systems, and it’s now the leading cause of failures in production agentic AI — systems that don’t just answer questions but take actions. Every agent you deploy is a new door.

What prompt injection actually is, in one page — a leave-behind that bridges engineer-to-executive translation.

Does anyone on your team know the difference between an AI chatbot that answers questions and an AI agent that can take actions on your systems — and which one you’ve actually deployed?

DoCRA is becoming the go-to methodology behind “reasonable security.” It forces a documented, harm-based justification for which AI controls get funded and which don’t, instead of a gut-feel best-practices list. Courts, regulators, and insurers increasingly expect to see that reasoning, not just a policy binder.

How to answer “did we do enough?” before anyone asks it — AI readiness as a repeatable methodology, not a feeling.

If your AI program were investigated tomorrow, could you show the math behind why you chose the controls you chose — or would it just be “this felt reasonable”?