The June 2026 Techeconomy Business Series took a deeper look at AI this time, less “How exciting is AI?” and more “What happens when someone tries to break it?”
The panel brought back familiar faces from the series: Peter Ndukwo, a Web3 security researcher and ZK contributor at Zipel Labs; Oluwatomi Obinna Alagbe, Staff Software Engineer at Malwarebytes; Abisola Rachael Aderohunmu, Product Manager Lead at Heala Tech; and Francis Udogu, a digital marketing manager.
The question on the table: how do you actually roll out autonomous automation without wrecking system security, consumer trust, or your own reputation in the process?
“They will find the loopholes you didn’t think of”
Peter Ndukwo started with a point: attackers don’t use your product as designed. They look for gaps you missed. Ndukwo has audited protocols like Chainlink and ZetaChain. He thinks companies rushing to add AI agents are making the mistakes that caused expensive hacks in DeFi.
Then it wasn’t usually a total system failure. It was one overlooked loophole that drained a protocol. Ndukwo criticized what he called implementation. Teams get seduced by an interface or fast demo but ignore vulnerabilities.
In Web3 that oversight means funds disappear permanently. Ndukwo’s argument was simple: if the system design is weak it doesn’t matter how good the product looks. It collapses when someone tries to break it. His fix is to limit agents to structured, logic.
Oluwatomi backed this up with the engineering side. He pointed to two attack paths: context poisoning and prompt injection.
Attackers can slip instructions into an agent’s context data. If that agent has access to databases someone can manipulate it into resetting passwords or exposing data.
Ndukwo and Oluwatomi made the point although from different lenses: full automation without strict isolation isn’t innovation. It’s a liability.
Francis Udogu wasn’t fully convinced. He thinks much caution slows companies down. For him speed and cost are key. AI should be used to scale outreach and predict customer needs.
Rachael landed somewhere in the middle. She agreed speed matters. Said successful AI projects solve specific problems for users. Chasing trends rarely works. Solving friction points does.
Ndukwo closed with a prediction. As AI models get cheaper and commoditized, advantages, like speed and efficiency stop meaning much.
The biggest competitive advantage left will be trust and identity. People will trust platforms they actually trust.
Automation should make systems more secure and reliable. It shouldn’t be an excuse to stop paying attention.
