I spent Wednesday morning at the fourth Safaricom Cybersecurity Summit in Nairobi, where the theme was “Harnessing AI, Strengthening Security”. Peter Ndegwa, Safaricom’s Group CEO, gave a speech that set the terms of the day. In a nutshell, he made the case that trust, not technology, is the scarce resource in Africa’s AI economy.

A CEO Who Has Already Been Deepfaked

Peter Ndegwa opened with a confession that a cloned version of his own voice had been used to advertise a fake Safaricom promotion. “Some of my team told me they would never have known it wasn’t me,” he said. Nicholas Mulila, Safaricom’s Chief Corporate Security Officer, had mentioned the same incident minutes earlier when he opened the summit. It is one thing to read about synthetic media in a World Economic Forum report. It is another to hear the CEO of a business with 57.9 million mobile subscribers in Kenya say it has already happened to him.

He also told us that cybersecurity has been the top risk on Safaricom’s risk matrix, by likelihood and severity, for the entire time he has been CEO. That is six and a half years. If the largest technology business in East Africa cannot get cybersecurity off the top of its register, nobody who was in that room should expect to do so either. 

Power Now Comes From Data, Computing & Algorithms

The basis of the speech came from António Guterres’ speech to the UN General Assembly on the 22nd of September 2026, where the Secretary-General argued that power once came from territory, then industry, then finance, and today increasingly comes from data, computing capacity and algorithms. Guterres named AI as one of four great tests facing humanity, alongside war, inequality and climate. Peter Ndegwa’s point in borrowing that point of view was very much on point. 

“The debate is no longer whether AI works, but how fast we should move and how firmly we should hold the reins.”

He added the line that should be the basis of every AI deployment in Kenya: “The risk is not AI itself, but AI used without governance, accountability or clear human oversight.”

Ndegwa shared that he has been to China twice this year and he described shipping ports that operate without a single human being in some processes. He was careful not to present that as the model for Africa. His view is that the continent’s opportunity lies in applying AI to financial services, healthcare, education, agriculture and public services, not in competing to build the largest AI model. “Africa does not have to be the consumer of AI. We can become one of its knowledge engines,” he said. 

150 Million M-PESA Transactions A Day 

In his speech, Peter Ndegwa said M-PESA now handles a mind boggling 150 million transactions a day. Safaricom’s FY2026 annual report, for the year to March 2026, put M-PESA at 46.86 billion transactions worth KShs 41.68 trillion, which works out at roughly 128 million a day across that year, so a 150 million daily run rate suggests the curve is still rising. 

Against that volume, Safaricom’s AI-driven behavioural models now stop about 94% of attempted fraud in real time, by learning customer patterns and where risk is coming from before the transaction completes. “There is no way we can anticipate and secure those customers without AI,” he said, and he was candid about why it matters commercially: “Cybersecurity is no longer an IT matter. It is a leadership responsibility, and it is increasingly a question of business growth, because customers will only adopt what they trust.” The platform that makes this possible is Fintech 2.0, the new M-PESA core launched in September 2025, which lifted capacity from 4,500 to 6,000 transactions per second with extra capacity to grow to 12,000. “Customer protection is not added at the end. It is designed into how we build,” Ndegwa said of it. He described the same philosophy running through Safaricom’s managed security operations centre, which it now sells to other organisations.

The 6% Problem

94% per cent is an impressive number. At 150 million transactions a day, the fraction of fraud that gets through is not a rounding error, it is a population. The Interpol African Cyberthreat Assessment Report 2026 says SIM swap fraud in Kenya surged by 327% in 2025, with more than 123,000 fraudulent SIMs issued and an estimated USD 3.8 million, roughly KShs 490 million, drained from mobile wallets. What is less often reported is that Interpol’s footnote for that 327 per cent figure is Safaricom’s own disclosure, in October 2025, that its SIM swap fraud investigations rose from 11 cases to 47. 

This is a reminder that in Kenya, Safaricom is the data. Peter Ndegwa set the standard himself when he defined what trust requires: that data is handled with respect, that systems stay resilient, “and that leaders are accountable when something goes wrong, and they do something about it.” The test of that sentence is already being written by the courts. Business Daily reported in July 2026 that DTB and Safaricom were ordered to pay a customer KShs 4.4 million over a fraud loss. Trust, in other words, is no longer only a brand promise. It is a liability line.

The Global Benchmark Kenyan Boards Should Measure Against

Peter Ndegwa cited the World Economic Forum’s Global Cybersecurity Outlook 2026, which found that the share of organizations with a process to assess the security of their AI tools nearly doubled in a year, from 37% in 2025 to 64% in 2026. He called that both encouraging and incomplete. If a third of organisations globally still deploy AI tools with no security assessment at all, the proportion in Kenyan boardrooms is almost certainly worse.

Only As Strong As The Weakest Link

“Africa’s AI future is not decided by who has the biggest model or data centre. It will be decided by who earns the most trust.”

He followed it with the reason no business, however large, gets to do this alone: “No single organisation can achieve this, because we are only as strong as the weakest link. We must work together to set AI standards, to share intelligence, to build capability, and to protect citizens without closing the door to progress.” He closed with quite possibly his most important statement from his speech spine: “Trust is what makes innovation last.”

Ultimately, What Does This All Mean? 

First, trust is earned at the margins, not in the averages. The 94% builds confidence; the handling of the remainder decides it. Second, the secure-by-design claim for Fintech 2.0 is the right posture because the platform is live and the fraud data will tell us over the next two years whether it holds. Third, Peter Ndegwa’s definition of Africa as a knowledge engine rather than a consumer of AI is only credible if the data that trains these systems is African, which is precisely the challenge Cabinet Secretary Hon. William Kabogo Gitau proposed later that morning. He closed by asking each of us to name one problem AI could solve in our organization and one step we would take to strengthen trust as we adopt it.