Corporate attempts to slow down artificial intelligence (AI) adoption in the name of governance may be backfiring, with employees increasingly turning to unsanctioned tools, exposing companies to data leaks, compliance failures and cyber risk.
This was the warning from Sibusiso Ngubeni, head of the data office for strategy enablement at Standard Bank Group’s CIB division, speaking at the ITWeb AI Summit on Wednesday.
Ngubeni said many companies face a dangerous mismatch: business units want AI's speed and productivity gains immediately, while governance and compliance functions move too slowly.
“The hidden cost of slow AI adoption is not that we miss revenue,” he said. “It is new forms of unmanaged risk.”
Ngubeni said the debate over whether AI belongs in the enterprise is over. “It already has [entered]. The real question is whether it enters through the front door with governance, or through the side door.”
That “side door” is shadow AI – the growing use of public or unapproved AI tools by employees seeking faster ways to work. “Shadow AI is just about employees trying to get their jobs done,” he said.
His comments reflect a growing concern among South African CIOs and CISOs that staff are feeding sensitive corporate information into public AI systems without approval.
Using psychologist Daniel Kahneman’s concept of fast and slow thinking, Ngubeni said the modern workforce expects instant answers. “Everybody demands to leverage AI now, and they don’t want to wait.”
But large organisations respond with slow approval processes, fragmented ownership and outdated governance, creating the perfect environment for unauthorised AI use. “If my current toolset doesn’t let me be productive, I will find another way,” he said.
Organisations that delay AI enablement often pay twice – first through lost productivity, then through incident response.
Rather than trying to suppress AI use, Ngubeni urged leaders to move faster with secure, sanctioned deployments. “The answer is not to slow down. It is to speed up, but safely.”
He recommended starting with two to three high-value use cases, supported by approved tools, clear accountability and strong data controls. “Centralise policy and governance, but federate adoption. Let people learn for themselves.”
Ngubeni said executives should stop viewing shadow AI as purely a disciplinary issue. “Shadow AI exists because there is unmet demand. We are not meeting business at their point of contention.”
Share
