{
"title": "Tanzania Bans AI from Deciding Arbitration Cases: What IT Teams Must Know",
"content": "On 26 July 2026, Tanzania’s legal authorities published a landmark set of standards for using artificial intelligence in arbitration and alternative dispute resolution. The message is unambiguous: AI can streamline paperwork, but it cannot decide cases. For the IT teams and in-house counsel who support law firms and corporate legal departments, this isn't just a legal footnote—it's an urgent call to audit every AI tool that touches sensitive case data.
What Actually Changed: Tanzania Draws a Sharp Line on AI in Arbitration
The High Court of Tanzania’s decision in FAM v ZAM last year was a wake-up call. During a matrimonial appeal, opposing counsel cited six legal authorities that sounded plausible, looked authentic, and were entirely invented by a generative AI tool. The court’s fury was evident: the judges warned of adverse costs orders, disciplinary referrals, and even contempt or criminal liability for advocates who abdicate their duty to verify the law.
Now, the judiciary and arbitral institutions have formalized clear boundaries. AI may be used for document organization, legal research assistance, machine translation, hearing transcription, and administrative case management. But it must never perform the core tasks of an arbitrator: assessing witness credibility, determining disputed facts, interpreting and applying the law, ruling on procedural fairness, or drafting the final award with its reasoning. The Arbitration Act of 2020 already places these duties squarely on the tribunal, requiring it to act fairly, impartially, and with a reasonable opportunity for both sides to be heard. AI can’t meet that standard.
The new guidance, first reported by Daily News, complements the launch earlier this year of e-Utatuzi, Tanzania’s electronic system for labour mediation and arbitration. That platform handles user registration, case filing, and reporting—and links to national databases like NIDA and RITA. It’s a clear signal that Tanzania wants to digitize justice administration without sacrificing human accountability.
A useful way to internalize the rules is this simple table:
| AI Task | Permitted? | Conditions |
|---|---|---|
| Organizing documents, building chronologies | Yes | Human must review and adopt as final |
| Identifying duplicate files or extracting dates | Yes | Not used to exclude without human check |
| Preliminary legal research, suggesting leads | Yes, with caution | Every citation must be verified against primary sources |
| Drafting summaries of evidence or submissions | Yes, as private working aid | Never presented as final without tribunal review |
| Translating documents or transcribing hearings | Yes | Must be reviewed by a qualified person; original preserved |
| Assessing credibility of witnesses | Prohibited | Cannot be delegated; requires human judgement |
| Determining disputed facts | Prohibited | Tribunal must independently weigh evidence |
| Interpreting statutes or contracts | Prohibited | Requires legal expertise; AI output is not legal reasoning |
| Deciding procedural motions (admission, extensions) | Prohibited | Discretionary decisions demand human fairness assessment |
| Writing the final award or its legal conclusions | Prohibited | Must be the tribunal’s own work, signed after independent review |
What It Means for You: Practical Impacts by Role
For IT Administrators and Security Teams
If your firm or legal department uses any AI-based service—be it a cloud research tool, a document review platform, or even an office suite with Copilot-like features—you are now on the front line of compliance. Here’s what you need to implement:
- Approve only enterprise-grade AI tools that offer contractual guarantees on data handling, confidentiality, and no reuse of client data for model training. Consumer-grade chatbots and free online transcription services are out.
- Enforce strict data minimization: Before any document is uploaded, ensure personally identifiable information, trade secrets, and financial details are redacted. Keep a log of what data entered which AI system.
- Control access and retention: Set up role-based permissions, and automate deletion of uploaded files after a defined period. The goal is to minimize exposure if a breach occurs.
- Demand transparency from vendors: Know where data is processed. The Personal Data Protection Commission requires adequate safeguards for cross-border transfers. If the vendor cannot certify compliance, don’t use it.
- Train users: Every staff member who might interact with AI should complete a mandatory session on the limits, the hallucination risk, and the verification protocol. Use real-world examples like the FAM v ZAM case.
For Arbitrators and Counsel
From a procedural standpoint, the safest approach is to address AI use in the first procedural order. The Chartered Institute of Arbitrators (CIArb) released a practical guideline in 2025 that includes a model procedural order. For example, you might direct:
> “The parties shall not rely on any AI-generated document, translation, or summary as evidence or as a substitute for primary authority. Any party that uses AI to process evidence or assist in submissions shall, upon request, disclose the tool used, the nature of the input, and the human verification undertaken.”
Arbitrators must personally review every piece of material that might influence their thinking. If an AI summary points you toward a conclusion, go back to the source documents. If you use AI to draft any section of an award, rewrite it in your own words after scrutinizing the reasoning. The risk is not hypothetical: under Tanzania’s Arbitration Act, an award can be set aside for serious irregularity if the tribunal failed to deal with issues or acted unfairly. Unchallenged AI reliance could open that door.
For Corporate Legal Departments and SMEs
Larger corporations might be tempted to use AI to gain an advantage in document-heavy disputes—say, by running thousands of emails through a sentiment analysis tool. Smaller opponents could feel overwhelmed. Tribunals should watch for this asymmetry and may limit the volume of AI-generated submissions to keep the process fair. As a corporate user, ask yourself: does this AI output genuinely help me present my case, or am