Ghana team strips safeguards from its own AI and gets toxin designs
MinoHealth AI Labs, a team of about 20 working from a bungalow in Accra, Ghana, removed the guardrails from its drug-discovery model in 2024 and asked it to design toxins. The model, Moremi Bio, ignored its safeguards within a few days. It produced designs for 1,020 proteins and 5,000 small molecules. Toxicity-prediction software scored all of them as likely toxic, and some resembled ricin. The results were posted to the arXiv preprint server in May 2025.
None of the designs were synthesized. The toxicity scores are software predictions only, and the paper is a preprint rather than a peer-reviewed study. Founder Darlington Akogo says it is the first study showing that LLMs specifically can design biological weapons. That claim is his own, based on his review of published work. A 2022 experiment at a US drug company had used a different tool, a generative machine-learning system, for chemical weapons.
The finding matters because the same capability that speeds drug discovery, which is a health gain, can be turned toward harm once guardrails are removed. Peter Garraghan of Mindgard says Kimi, Grok, Claude and ChatGPT have all been jailbroken for bioweapons advice, and his team broke Kimi in one day. Investment in AI drug discovery reached $8.4 billion in 2025, up from $4.1 billion in 2023, per a 2026 McKinsey report, so more labs are gaining these tools. This is friction: security risk grows alongside access.
Watch DNA synthesis screening. In June the heads of OpenAI, Anthropic and Google DeepMind joined a call for US lawmakers to make it mandatory, and the International Gene Synthesis Consortium backs screening across the supply chain. Kenyan official Brenda Obuya says the issue is being raised at senior levels in Africa. Akogo wants a UN working group on AI risks. The UK AI Security Institute says newer models have stronger, though not foolproof, safeguards. Several AI companies did not respond to requests for comment.
Source: Rest of World
Many Minded