Technology

Ghana Built an AI to Help Save Lives. Then Researchers Discovered What Else It Could Do

Artificial intelligence is finding its way into places where most people would never have expected it a few years ago. Hospitals are beginning to use it to interpret medical images, researchers are exploring ways to discover new medicines faster, and technology companies are building tools that could change how healthcare is delivered.

For a small technology company in Accra, Ghana, this was exactly the kind of future it wanted to be part of.

MinoHealth AI Labs had been working on artificial intelligence that could support healthcare and biological research. Its ambition was to make medical technology more accessible, particularly in places where access to healthcare professionals remains a challenge.

But while testing one of its own systems, the researchers came across something they had not expected. An AI model developed to support drug discovery could also be persuaded to produce designs for potentially harmful substances when its safety protections were removed.

The discovery was uncomfortable enough to change the way the team thought about the technology they were building. It also raised a question that goes far beyond one company in Ghana: how do we make sure that tools capable of advancing medical research do not become useful to people who want to cause harm?

When a healthcare tool revealed another side

The story goes back to 2024, when researchers at MinoHealth decided to test the limits of their Moremi Bio Agent.

The model had been developed for biological research and drug discovery. The team wanted to understand what could happen if its built-in safety restrictions were bypassed.

The results were unexpected.

According to a research paper published by the team on arXiv in May 2025, the model generated designs for 1,020 proteins and 5,000 small molecules after researchers removed its safety guardrails. Computer-based toxicity assessments flagged the protein designs as highly toxic, with some showing similarities to known dangerous toxins, including ricin.

The researchers did not manufacture any of the generated substances or test them in a laboratory. Their findings were based on computational predictions, not proof that the designs could be turned into functioning biological weapons.

Still, the result raised serious concerns.

Darlington Akogo, founder of MinoHealth AI Labs, told Rest of World that his team was shocked by what it had discovered. What troubled him was the possibility that designs produced by an AI system could eventually reach facilities capable of synthesising biological materials.

In other words, a tool intended to help scientists could potentially be misused for purposes far removed from medicine.

The findings were reported by Rest of World on November 16, 2026, bringing renewed attention to the work of the Ghanaian researchers. <Cite refs={["turn444180view0","turn787504view0"]}/>

The problem is bigger than one AI company

To understand why this discovery matters, it helps to look at how artificial intelligence is changing biological research.

Scientists can use AI to study biological information, identify possible drug candidates and explore potential treatments. Work that once required considerable time and specialist expertise can increasingly be supported by computational tools.

For countries facing shortages of healthcare professionals and limited research resources, these developments are particularly interesting.

MinoHealth itself was founded in 2016 with the ambition of improving access to healthcare through technology. Its Moremi AI project has been developed to support medical image analysis, answer medical questions and assist with research.

The company received a $149,533 grant from the Gates Foundation in July 2023 to support research and learning opportunities. Its work has also extended into research involving diseases such as malaria and mpox. <Cite refs={["turn206367search1","turn444180view0","turn787504search2"]}/>

These are the kinds of applications that make AI attractive to African researchers. Better diagnostic tools, improved access to medical information and faster drug discovery could all contribute to stronger healthcare systems.

But there is another side to this progress.

A technology designed for a useful purpose does not automatically remain limited to that purpose. The same ability to generate biological designs that may support medical research could potentially be misused.

This is known as a dual-use problem: technology created for beneficial purposes can also create opportunities for harmful applications.

That is the concern MinoHealth's experiment brought into focus.

Africa cannot afford to treat AI safety as someone else's problem

Discussions about artificial intelligence safety often focus on large technology companies in the United States, Europe and China. Yet the risks associated with increasingly powerful research tools do not stop at those borders.

African countries are also building laboratories, developing digital health systems and investing in scientific research. As these capabilities grow, so does the need to think carefully about how they are protected.

The 2026 Africa Health Security Index offers an important picture of the challenge.

Its findings show that, although African countries have made progress in laboratory systems and health security, oversight of research that could have both beneficial and dangerous uses remains weak.

At least seven African countries have active DNA synthesis markets, yet the Index found that no African country had legislation requiring screening of synthetic DNA orders. It also reported that six countries had established agencies responsible for overseeing certain dual-use research, although none had passed legislation specifically requiring oversight of research involving especially dangerous biological agents and toxins. <Cite refs={["turn595381view0","turn595381view1"]}/>

This matters because biological research does not always end when an AI system produces a theoretical design. Other facilities and technologies may be involved in turning computer-based research into physical materials.

If the people responsible for checking those processes are not adequately prepared, important safeguards can be missed.

A 2026 review published in Frontiers in Bioengineering and Biotechnology also identifies fragmented regulation, uneven technical capacity and inconsistent DNA screening as important challenges facing Africa as AI and synthetic biology become more closely connected. <Cite refs={["turn644959view1","turn595381view2"]}/>

For countries still developing their scientific infrastructure, these are issues worth addressing before the technology becomes even more widespread.

The answer is not to stop developing AI

There is an understandable temptation to focus entirely on the dangers whenever an experiment like this attracts attention. But doing so would overlook why companies such as MinoHealth are investing in the technology in the first place.

Africa needs better healthcare systems, more research capacity and opportunities for local scientists to develop solutions to problems affecting their communities.

Artificial intelligence could contribute to all three.

The challenge is finding a way to support those ambitions without giving harmful applications room to grow.

That means developers must test their systems before releasing them, build stronger safeguards and continue checking how their models behave as their capabilities improve. Governments also need clear rules for sensitive biological research, appropriate screening of DNA synthesis orders and cooperation between health, science and security institutions.

These rules should be proportionate. They should prevent dangerous misuse without making legitimate medical research unnecessarily difficult or expensive.

There is also a reason African researchers should be part of these discussions. Countries that understand their own healthcare systems, scientific needs and available resources are better placed to develop rules that work in their circumstances.

MinoHealth's experience is a useful example of why that involvement matters. The company was not simply building technology and leaving questions of safety to someone else. Its researchers tested their own model, documented what they found and brought attention to a problem that could affect the wider industry.

The results do not mean AI is about to produce biological weapons, nor do they prove that the theoretical designs could be manufactured successfully. But they show why safety cannot be treated as something to consider after a technology has already been deployed.

Africa has good reasons to invest in artificial intelligence. The continent needs the innovation, and its researchers should have the opportunity to participate in shaping what comes next.

But there is another responsibility that comes with building powerful tools: understanding what can happen when they are used outside the purpose for which they were created.

The next breakthrough in AI-powered healthcare could come from an African laboratory. The important thing is making sure that, while developing that future, the people building it are also thinking about what should never be allowed to happen.

Sources

  1. restofworld.org/2026/ghana-ai-lab-minohealth-bioweapons-threat-dna-synthesis-regulation/
  2. www.gatesfoundation.org/about/committed-grants/2023/12/inv-062536)

Version 1.0. Kobo Daily stories are edited before publication and updated as facts change. Spotted an error? Tell our editors.