AI Standards
Essays
On how intelligent systems should behave, where they fail, and who answers for them.
- SecurityShadow AI: discovery and controlShadow AI is deployer use you have not listed. Hunt with the sensors you have. Bin every finding. Do not invent a percentage.
- LeadershipWhat a board should ask about AIFifteen questions a director can ask without a demo. If the answers need adjectives, the company does not govern AI yet.
- SecurityPrompt injection: a CISO playbookInjection is the method. The tool budget is the cause. Six controls you can show, and a test that measures tool calls — not manners.
- AI SystemsWhen to run models on hardware you ownThree tests: leak, swap, outage. Fail one, and a hosted model is a concentration. Pass all three, and an API is still allowed.
- SecurityModel provenance and weight hashingModel cards can be copied. Weights can be swapped. Hash what you load, pin it, and log the digest — or admit you cannot name what ran.
- GovernanceISO/IEC 42001 explained for the board42001 is how you govern AI, not how you bless a model. Read the scope. Keep the Act on the calendar.
- GovernanceWho is liable when an AI agent acts?An agent has no legal personality. Provider, deployer, product liability, and ordinary agency decide who pays — and a jailbreak does not change that.
- RegulationWhat the EU AI Act requires in 2026–2027The Act is already live. High-risk duties moved. Here is what each seat owes now, and what 2027–2028 actually changed.
- Sovereign AIBuilding Sovereign AI: The Distance Between a Vision and a Working ProductSovereignty is not branding. It is a set of operational decisions and tradeoffs.
- ArchitectureThe Architecture of an AI That Knows What It Doesn't KnowGood AI architecture makes uncertainty visible and enforceable.
- AI GovernanceA Blockchain Can Preserve a Wrong Answer ForeverImmutability establishes a record. It does not establish truth.
- AI GovernanceThe Problem With AI That Always Says YesAn assistant that refuses to acknowledge limits is not helpful. It is dangerous to depend on.
- AI SystemsWhy I Started Building AI Systems Instead of Just Using ThemUsing AI is easy. Making it dependable in a real workflow is another discipline.
- EnterpriseWhy Most AI Implementations Fail (And How to Fix It)After 17 years in enterprise systems, I've seen the same patterns destroy AI projects over and over. Here's what actually works.