Practitioner insights on AI governance, operational compliance, and responsible AI.
For twenty-five years the UK Post Office prosecuted more than 700 subpostmasters on the say-so of its Horizon accounting system. Not AI — included for the lesson. The gate is that a system's output accusing a person is independently verifiable, and is not legal evidence by itself. Level 400, the audit gate.
Read more →
Australia's automated debt scheme averaged annual income into fortnights, fabricated debts for anyone with irregular earnings, and reversed the burden of proof onto about 440,000 people. The gate is not on the notice. It is on the arithmetic, before a single notice goes out. Level 400.
Read more →
ProPublica reported that Cigna's PXDX process let medical directors deny large batches of claims in seconds each, often without opening the patient's file. The gate is a human who actually reviews before a denial — and the arithmetic of what "actually" means when the batch is a thousand deep. Level 300.
Read more →
In 2012 a faulty deploy sent $7 billion of unintended orders in 45 minutes and cost Knight Capital $440 million and the firm. No LLM anywhere. The gate it needed is the one your agent needs today: the deploy is a consequential action, and it gets a rule, a kill switch and a rehearsed way back.
Read more →
In 2023 a dealer's chatbot agreed to sell a 2024 Chevy Tahoe for $1 and called it a legally binding offer. Nobody lost a truck. Everybody learned that a polite instruction to an LLM is not a gate. Level 200 starts here.
Read more →
The Foundation's home page is a café now. Wren keeps the table by the door and asks one question. Five doors, each a real case; ten minutes to start; an hour to a rule in your own words, a signed receipt, and a Thursday.
Read more →
The Foundation's open-source projects become one practice with four levels, each an hour or less, each on a real case from the 100 verified harms — built on OVERT, the open standard from Glacis, which is the hero of this story.
Read more →
Ten yes/no questions a manager can answer about an AI agent this week, each with the evidence to point at, ending in a card you keep. No account, no maybe.
Read more →
ISO 42001 is the AI management system standard. Treat its controls as checks you run in CI, each leaving a receipt an auditor can read, not a document you file.
Read more →
Write for the LLM that's actually reading you. Your CEO is not reading your 60-page strategy doc. Their LLM might.
Read more →
We have officially moved beyond the era of "AI as a tool" into the era of "AI as an employee."
Read more →
The inaugural AiGovOps Foundation semi-annual Healthcare AI Governance report. Where the field actually stands.
Read more →
Every decade we rediscover a new flavor of the same old problem. Meet Agentic Hell — recursive agent loops with no observability.
Read more →
A practitioner's guide to the modern AI Ops stack and how to align governance with deployment pipelines.
Read more →
Every AI system deployed without automated governance creates technical debt that compounds. Here's what we're doing about it.
Read more →
88% of organizations are using AI, but only 39% have board-level oversight. That's not a gap — that's a canyon.
Read more →
August 2, 2026 marks the applicability date for key EU AI Act obligations. Practical clarity on what you need to do now.
Read more →
AI agents are the backbone of the global economy. But a dangerous phenomenon has emerged: Anonymous Ghosts.
Read more →