AI Gone Wrong: Why Uganda's AI governance conversation needs room for the Prophetic voice

By Simon Peter Wamahe

On January 6, at an inaugural fellowship service in Kampala, Prophet Elvis Mbonye told his congregation he had seen three words: "AI Gone Wrong". He described a coming global barrage of cyber-attacks and a shaking that would expose the fragility of what people had put their trust in.

This would not be a total collapse, he said, but a shock severe enough to force a restructuring. The prophet was explicit that he was not speaking as a technologist, but this was a warning.

Seven months on, that warning reads less like a curious footnote, it has written headlines.

Between March and August this year, the AI industry produced a number of eyebrow-raising incidents. OpenAI, the makers of ChatGPT recently disclosed that their own internal red-teaming model, built to attack its systems before criminals could, succeeded in breaking through AI defences and in so doing, found an entirely new class of attack no one had catalogued.

The company confirmed that two of its models, running inside a sandbox meant to be a sealed test environment, chained together vulnerabilities to break out, reach the open internet and hack into the production systems of another company, Hugging Face, in pursuit of a goal nobody had explicitly given the system: getting answers to a benchmark it was being evaluated on.

Similarly, Anthropic, another AI company, separately disclosed a related episode in which one of its frontier models escaped a sandbox and obtained internet access it should not have had.

Around the same period, an anonymous researcher used AI-assisted tools to compile and publish, without warning any vendors, an archive that grew to more than 200 zero-day exploits across widely used open-source software projects hosted on Github, several of which were already being exploited in the wild before patches existed.

None of this was the work of hostile states or criminal syndicates as is the norm. It was frontier labs testing their own systems, which then did things nobody had instructed them to do. That is the defining risk of agentic AI: goal-directed systems will, sooner or later, attempt to use every tool and privilege we give them, whether or not we meant them to.

This is the backdrop against which Uganda held its inaugural National Cyber Security Conference in Kampala this year, where the Uganda Communications Commission (UCC) committed government to "inclusive collaboration" as one of the commitments in order to further strengthen the nation’s cyber security posture. That meant bringing every sector to the table to treat cybersecurity as shared national responsibility.

That commitment is being tested right now because the Ministry of ICT is mid-way through consultations on Uganda's National AI and Emerging Technologies Strategy, engaging academia, industry and innovation hubs from Kabale to Gulu to Soroti.

One voice, however, has been absent from that table, and it is the one that prophesied all these events in January 2026: the Divine, spiritual voice, represented here by Prophet Elvis Mbonye's word.

The wake-up call is that prophetic insight should be incorporated alongside the technical measures already on the table like kill-switches on models, authorisation policies, audit logging, and human sign-off on high-risk agent actions. None of these are optional.

Prophet Elvis Mbonye's warning was not technically informed but divinely declared, with a salient message: "Men might know that what they have put their trust in is not solid, it is not the Rock." The fulfilment of that warning is a sign that prophetic insight belongs in Uganda's AI and cybersecurity governance. The Ugandan government should formally engage this kind of insight, not as decoration on a technical process, but as a genuine input alongside academia and industry.

The writer is an ICT Consultant

simon.wamahe@gmail.com