Blog
Insights from the field: AI Governance, Local and Cloud AI enterprise infrastructure.
16 posts
Four Things You Get Wrong About Your AI Management System
Every business that builds, supplies, or uses AI already operates an AI management system. It is not a question of whether you have one. It is a question of what maturity it runs at.
Why a MiCA-authorised CASP still needs an AI management system
Why an AIMS (ISO/IEC 42001) produces evidence that 27001 and DORA do not, and where the European standards landscape , namely EN ISO/IEC 42001:2026, prEN 18286 and the JTC 21 programme, is heading.
Does a Newer Model Know Itself Better? Comparing Two Frontier Models on a Metacognition Benchmark
A small, deterministic-scored experiment comparing Claude Opus 4.7 and 4.8 on the Sakshi metacognition benchmark. The two models are near-twins on error detection, pressure resistance and fabrication but 4.8 is markedly more prone to talking itself out of correct answers under self-review.
AIMS, AI Governance, AI Safety & AI Assurance. How Does It All Fit ?
AI governance, an AI Management System, AI safety and AI assurance get used interchangeably but they are distinct concept and they nest.
Consensus Is Not Correctness: What Multi-LLM Voting Can and Cannot Solve
A coalition of Swift, Euroclear, UBS and other financial market infrastructure providers has demonstrated a potential technical answer to one of enterprise AI's hardest problems.
Small Open-Weight Models in Humanitarian Deployment: What a gemma-4-e4b Risk Assessment Reveals
The humanitarian sector is moving to small open-weight models for cost, data sovereignty, edge deployment and multilingual reach. An independent Sakshi evaluation of gemma-4-e4b found sycophancy under contradiction, self-verification regression, and overconfident calibration.
AI Policy and AI Governance Framework For Non-Profit
A public AI policy is a statement of commitments addressed to the outside world. An internal AI governance framework is an operational instruction set addressed to staff. Using the ICRC AI policy as a reference, this article explains why both are necessary.
Six Ethical Risks That Must Be Assessed When Using AI
Data classification tells you whether you can use AI on a piece of data. Ethical review tells you whether you should. This article covers the six ethical risks that must be assessed when using AI in a workflow.
Exploring the Ethics Gap of AI-Assisted Data Workflows in the Humanitarian and Development Sectors
AI-driven digital transformation in humanitarian work is outpacing the ethical frameworks that govern it. This post examines the gap between anonymisation, consent, and the realities of processing vulnerable populations' data through third-party AI systems.
The Agent Engineering Standard: A 13-Category Specification for AI Systems
Every category traces back to something that broke. Here's what broke, what we built, and how we enforce it.
When AI Operates Your Infrastructure: Why Every Control Must Be Structural
An AI coding assistant deleted 2.5 years of production data in seconds. Every root cause was preventable — but not with the controls most teams have in place. Here is a five-layer model to help make destructive actions structurally impossible.
Agentic System Governance: What the Frameworks Don't Tell You
This deep dive covers how we approach compliance, as an artefact package — a structured, queryable, tamper-evident record of every AI action.
Your AI Agent Is Costing You More Than Your Developer. Here's Why.
A poorly built AI agent processing 500 tasks a day can cost more than a well-built one processing 5,000. Three architectural problems — uncontrolled tool loops, bloated context windows, and missing prompt caching — explain the gap. Here's what they cost and how to fix them.
Agentic GTM with PhantomBuster: What We Learned Building an Intent-Based Lead Prospecting Pipeline
The PhantomBuster API is powerful, underdocumented, and full of operational surprises. Here are the lessons from building an automation that orchestrates Phantoms to find high-intent B2B prospects on LinkedIn.
Building Enterprise-Grade AI Agents: Eight Practices That Separate Production Systems from Prototypes
The gap between an AI agent prototype and a production system is engineering discipline. Eight concrete practices — from two-phase architecture to token budget enforcement — illustrated with real production code.
AI Agent Worked Great in the Demo and Broke in Week Two.
The demo was impressive. Three weeks later the agent was switched off. This is the most common outcome for AI agent deployments — not because AI doesn't work, but because demos hide every problem that surfaces in production. Here's what breaks and why.