The compliance library

The Evidence Room

6 · Read the guide

Evidence · Research

Enterprise AI can’t do your arithmetic — LG measured it (4%). Here’s the fix.

A 2025 study on LG Electronics’ real internal data: 93% accuracy on simple questions, 4% on arithmetic reasoning. Why improvisation fails in production — and the structural fix.

Read the guide →
Evidence · IBM

IBM just published its agentic-AI governance blueprint. Here’s what it means for the Gulf.

CUGA puts policy enforcement, human approval gates, and output control inside enterprise agents. The doctrine converges; sovereignty, Arabic, and GCC fit are the difference.

Read the guide →
Evidence · Market

The $320B question nobody is answering: who governs what AI does?

PwC projects $320B for Middle East AI by 2030. Markets pay ~56× sales for governed AI and ~4× for ungoverned automation. The money has voted: governance is the product.

Read the guide →
Evidence · BI reality

Your dashboard answers yesterday’s questions. Who answers tomorrow’s?

Even data-mature companies wait 10–14 days for a new question to become a report. Dashboards are viewing tools; business is an asking activity.

Read the guide →
Evidence · EU AI Act

The AI Act doesn’t require self-hosting. So why does everyone audit like it does?

High-risk duties deferred to 2027–28; Article 50 transparency was not. Why owning the environment is the cheap way to prove logging and oversight.

Read the guide →
Evidence · Red team

We red-teamed our own AI layer 63 times. Here’s the scoreboard.

63 attacks in Arabic and English across five database shapes: what broke, what held, and why we publish it. Scoreboards only — the proof culture, not the internals.

Read the guide →