Skip to content

AI & Automation Policy

How MaishaThabiti uses AI and automation while keeping evidence, validation, provenance and editorial accountability in control.

MaishaThabiti may use artificial intelligence and automation to increase research capacity, consistency and maintainability while keeping evidence, provenance and editorial accountability separate from machine fluency.

The core principle

AI is an assistant and interface, not the authority. A generated sentence does not become true because it is well written. The authority for an important factual claim should come from suitable evidence, and the authority for publication should come from the platform’s validation and review process.

Where AI can help

  • Discover candidate subjects and sources for research.
  • Classify and structure information into defined knowledge types.
  • Extract candidate facts from source material for review.
  • Compare records, identify contradictions and flag missing evidence.
  • Draft summaries or explanations grounded in approved knowledge.
  • Run consistency, schema and quality checks across large datasets.
  • Help users phrase questions and understand decision outputs.

What AI must not establish on its own

AI should not invent sources, silently fill unknown fields, create unsupported factual certainty, treat its own prior output as evidence or publish a high-impact conclusion solely because the model finds it plausible.

Draft-first governance

Automated research output should be treated as candidate or draft knowledge until it satisfies the relevant evidence, schema and review rules. The system is designed so that incomplete knowledge can remain incomplete instead of forcing the model to fabricate a value merely to satisfy a field.

Decision support

Where MaishaThabiti provides recommendations or decision outputs, the reasoning should be traceable to explicit user inputs and suitable, dated Knowledge Records. AI may explain the reasoning in natural language, but it should not independently choose unsupported facts or silently alter deterministic decision rules.

Auditability and reversibility

Automated workflows should preserve provenance and respect data boundaries. Material changes should be reviewable and, where the system supports it, reversible. A system that can publish unsupported certainty at scale is not an improvement simply because it is faster.

User-facing AI

If conversational or personalised AI features are added, users should still be able to distinguish sourced knowledge from generated explanation. High-stakes answers should encourage verification when the relevant facts are time-sensitive, incomplete or regulated.

The objective

Use automation to make MaishaThabiti easier to improve without making it easier to manufacture confidence. Scale should increase the amount of governed knowledge, not the amount of unsupported text.

Last updated: 6 September 2026.