OnTrack with eBlocks Software. South African organisations are eager to take advantage of AI, but concerns about ethics, governance, compliance and security are slowing adoption by many of them, says Deon Thomas, Managing Director of eBlocks Software, a software engineering specialist focusing on data, AI and DevSecOps.Thomas says AI has the potential to change the balance of power in business. “Smaller and mid-sized companies are moving to accelerate the use of AI to gain a competitive advantage and displace bigger competitors. But large enterprises – particularly in highly regulated sectors – are being held back by ‘cultural inertia’. “A large bank, for example, is going to be slower to embed AI into their system because of governance, security and compliance,” he says.“eBlocks has been implementing AI solutions for a number of customers. Without fail, every customer that we speak to wants AI, but almost all of them are concerned about AI ethics and how to implement responsible AI. So it’s very clear that customers need a framework that gives them confidence that they are implementing AI responsibly, ethically and securely,” Thomas says.A key concern customers cite is IP leakage. “They don't want to put their customers' personal information or their proprietary IP into these public AI models. Most of these agentic tools have plugins that authorise the AI to go into data unsupervised,” he says. “Organisations are also concerned about bias detection, fairness and transparency or explainability. They need to know if AI made a decision on something, where's the explanation for its decision and its logic? They want to know what controls were in place.“Many don't want to allow the AI agents to be too autonomous – they still want to keep a human in the loop because AI is not always 100% accurate,” Thomas says.These are valid concerns, he says. “For example, when a bank is looking at financial risk models, credit scoring can become inaccurate over time and if you don't maintain the AI models, they could make decisions or predictions based on stale data. Without monitoring, the right governance in place and humans in the loop, you could end up with incorrect decisions that create a compliance risk. In healthcare, if diagnostic tools are trained using poor quality data, you could miss critical symptoms and put patient safety at risk,” he notes. Thomas says successful, responsible AI requires an ethical framework, risk assessment and mitigation, data governance and security, and continuous monitoring and accountability.Thomas recommends starting with a proper risk assessment and an AI policy, with clarity on who is accountable for responsible AI within the organisation.“Accountability cuts across executive roles. For example, a chief data officer will look after data governance, a chief risk officer looks after security, and a compliance officer looks after audits. Currently, most organisations don’t have a particular person who's carrying the hat to ensure that AI guardrails are implemented properly. So we recommend that they always start by putting an AI policy in place and determining who should be accountable for making sure it gets enforced.”Thomas explains that AI ethics refers to the principles and moral considerations that would guide how AI is developed. “For me, AI ethics is the why – why are we doing this? Responsible AI is the how. Together they allow us to build trust into systems. It’s about how it's developed, how it is deployed and how it is actually used. This is important because it focuses on fairness, transparency and accountability,” he says.He concedes that the process of putting the right AI ethics and governance frameworks in place could slow down AI roll-outs somewhat, but says: “The risk of adopting AI without ethics is too big. But once organisations have the right guardrails in place, they will be able to move a lot quicker, with much less risk.”Thomas adds that training employees on the responsible use of AI is also important in building trust into intelligent systems. He says: “AI fails when the people who are operating it do not fully understand its capabilities or risks. Most of the large language models are prompt driven, so if an employee’s prompt is incorrect, they could get the wrong output.”To support customers in their AI journey, eBlocks has pioneered its AI-first software engineering transformation service – OnTrack – with a framework to expedite ethical, secure and trusted AI.Thomas says the eBlocks ethical AI framework rests on five key pillars.“Our framework is designed to make sure that we must empower organisations responsibly, while protecting people, data and trust. The first pillar is governance by design, meaning we embed responsible AI practices into the delivery process from the point of building the solution. The second pillar is transparency and explainability. We ensure transparency in the AI’s decision rules, with dashboards, monitoring, reporting and fairness,” he says. “Third is human-centricity and fairness, making sure that we prioritise AI that uplifts people rather than replacing them. Fourth is continuous oversight and improvement. After a ‘co-delivery’ implementation, we offer a managed service to monitor and manage the solution, refine the guardrails and make sure that there's no misbehaviour of any implemented systems or models. And the final pillar is the ethical use of data.”These factors are crucial for successful AI, he says. “We see many organisations rushing to implement AI without doing it in an ethical way. Finding the right partner to help with this allows them to move faster, with all the right pillars in place.”Read more about the eBlocks ethical AI framework here.