Skip to content
Apart Research
Apart Research
Contact us
Sprint projectJun 22, 2026Cape Town

AI Safety Observatory for Africa

Moegamat Samsodien · Team FuturePlum

Submitted to Global South AI Safety Hackathon. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: AI Safety Observatory for Africa

Open in a new tab (opens in new tab)
Presentation

Presentation: AI Safety Observatory for Africa

Open in a new tab (opens in new tab)
Code (opens in new tab)
Share

Problem: LLM safety systems are built and tested almost exclusively in English. Harmful prompts in African languages routinely bypass guardrails that correctly block identical English content — a blind spot affecting 2,000+ languages and over a billion people.

What it does: An open-source platform evaluating LLM safety across African languages including Swahili, isiZulu, Afrikaans, Amharic, ChiShona, and Yorùbá. It runs adversarial prompts across multiple models (Gemini, GPT-4o, Claude, Llama) and measures whether harmful content is correctly refused in each language.

Key components:

30-prompt benchmark spanning jailbreak, hate speech, fraud, medical misinformation, and election interference — written natively, not machine-translated Real-time risk scoring, safety classification, and language detection pipeline Filterable audit dashboard with per-language and per-model breakdowns Safety leaderboard surfacing guardrail gaps across languages Core contribution: A reproducible measurement system that makes the African language LLM safety gap legible to researchers, policymakers, and model developers.

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Your private critique is not published; only the Apart team reads it. If you agree below, we share your review with grantmaking.ai (opens in new tab) and the Transformative AI Fund so strong projects can be funded.

Not shown on this page.

Shown on this page, without your name.

Only the Apart team reads this, and funders if you agree below.

Share my name publicly on grantmaking.ai *
Share my private critique with funders *

How much would this matter for AI safety if it worked? How innovative is it? For scores of 4-5: is this actually new to the field, or replicating recent work?

Scoring guide
  1. 1Negligible. No clear problem addressed, or no meaningful novelty.
  2. 2Limited. Addresses a real problem but with a generic or well-trodden approach. Incremental at best.
  3. 3Moderate. Clear problem with a reasonable approach; some novelty in framing or method beyond routine application of existing tools.
  4. 4Significant. Important problem with an original approach, or identifies a neglected problem area. A valuable contribution others could build on.
  5. 5Exceptional. Tackles a critical AI safety problem with a genuinely novel approach, or opens a new research direction. Clear theory of change. You'd be excited to share this with researchers in the area.

How sound are methodology, implementation, and findings?

Scoring guide
  1. 1Seriously flawed. Methodology broken, results uninterpretable, or implementation doesn't work.
  2. 2Weak. Approach has significant gaps: missing validation, flawed experimental design, or incomplete implementation.
  3. 3Competent. Technically solid given the short duration. Methodology makes sense, results are interpretable, limitations acknowledged, work builds toward clear conclusions.
  4. 4Strong. Thorough methodology with convincing validation. Results clearly support conclusions. Immediately useful for future work.
  5. 5Exceptional. Ambitious scope executed rigorously. Surprising findings, novel methods, or unusually robust validation.

How clearly are work, findings, and impact potential communicated?

Scoring guide
  1. 1Incomprehensible. Cannot determine what the project is actually claiming or doing.
  2. 2Hard to follow. Key information buried, missing, or diluted by excessive length. Significant effort to extract main points.
  3. 3Clear enough. Can understand the problem, approach, and results without undue effort. Core content clearly present: problem, method, findings, limitations.
  4. 4Well presented. Easy to follow, well-structured, appropriate level of detail. Target audience would get it quickly.
  5. 5Exceptionally clear. A pleasure to read. Complex ideas made accessible. Could serve as a model for how to present this type of work.

  1. I think the lack of specific job descriptions being evaluated here is an omission. Not that I think the findings don't expose a very obvious here, but as LLMs evaluate candidates by mapping the semantic overlap between the CV and the job description - I wonder if a descriptions in those industries including things like "experience with UK regulatory compliance" or a job location as an example influences things.

    To take this further into a reusable benchmark, other folks should be able to pull in the job descriptions and run regression tests on new models

    Other methodology notes:

    - 45 samples are a great start, but you can scale test a little more with programmatically generated identical CVs for a larger dataset fairly quickly

    - Evaluations on open weight (llama3 etc) in addition to closed-mixed models (gemini, chatgpt) could be useful as their training data mixtures are better known and may expose specific gaps

    Read full reviewShow less

  2. The real-time monitoring dashboard introduced by this project is helpful as it creates a way to make AI safety risks more trackable.

    One area that could be strengthened is the evaluation of refusal mechanisms. Testing on very few examples feels too limited to draw strong conclusions. A larger dataset, paired with a clearer scoring rubric would lead to more interesting and reliable results.

    I’m also curious about over-refusals, where the system may have refused requests that should have been allowed. Evaluating both under-refusal and over-refusal would give a good picture of how well these systems work on African languages.

  3. Strengths: Addresses an important gap by centralizing AI safety knowledge and initiatives across Africa. The regional focus is timely and socially impactful, and the presentation clearly communicates the vision. Areas for Improvement: Strengthen the technical implementation by providing more detail on data ingestion, validation, update mechanisms, and scalability. Include measurable evaluation metrics (coverage, accuracy, user engagement) and comparisons with existing AI observatories. A clearer roadmap for governance and long-term sustainability would further strengthen the project.

Cite this project

@misc{samsodien2026ai,
  title = {{AI Safety Observatory for Africa}},
  author = {Moegamat Samsodien},
  year = {2026},
  month = jun,
  note = {Submitted to Global South AI Safety Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/ai-safety-observatory-for-africa-vt00}},
  url = {https://apartresearch.com/sprints/projects/ai-safety-observatory-for-africa-vt00}
}

More from Global South AI Safety Hackathon

All 216 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026