EY refers to the global organisation, and may refer to one or more, of the member firms of Ernst & Young Global Limited, each of which is a separate legal entity. Ernst & Young Global Limited, a UK company limited by guarantee, does not provide services to clients.
Recent Searches
Responsible AI cannot be installed like software; it must be embedded into culture, behaviours and workforce decisions.
In brief
- Responsible AI cannot be treated as a software deployment or compliance exercise. It is a workforce capability that must be continuously developed through learning, judgement and experience.
- "Human in the loop" governance only works when employees understand the AI outputs they are reviewing, validating and approving.
- Citizen-developed AI agents are becoming widespread, shifting risk from centrally managed platforms to employees who may lack training in data, validation, risk and decision-making.
Organisations are treating AI like software, not a capability. ‘Human in the loop’ only works if the human knows what they’re doing. Right now, many don’t – and that sits squarely on the shoulders of the CHRO.
Most people can’t do long division anymore. That doesn’t matter until you ask them to check a calculation.
This same dynamic is emerging with AI. Employees are being asked to validate outputs, override recommendations and take accountability for decisions produced by systems they don’t fully understand.
If Australia wants to be a trusted place to deploy AI, then trust is won or lost in the capability of people asked to make sense of AI outputs.
Just 9% of Chief Human Resources Officers (CHROs) can identify the appropriate controls for key AI risks. This is a signal that risk is moving faster than traditional people frameworks.
The latest EY Responsible AI Pulse Survey shows organisations are reporting strong gains in productivity, speed and innovation. But it also shows just 30% of HR teams have started developing strategies for managing a hybrid AI-human workforce.
Many organisations are approaching AI as a one-off technology rollout. Policies are written and people sign off forms, but this is like asking people to read the terms and conditions on their phone upgrade. Box ticked and forgotten.
The riskiest activity in organisations right now is not centrally approved AI platforms. It’s citizen developers building their own agents on un-approved and un-managed AI platforms. Two-thirds of organisations now allow employees to do this, according to the survey. And why not? Attempts to block this activity rarely work. People find a way around the controls that are being relied upon.
We can’t install responsible AI. Rather than a principle, it is an ongoing process of awareness, training and skills development that evolves with the workforce.
Building an AI agent safely requires far more than access to low-code, no-code technology and some technical curiosity. It demands judgement about data, process, validation, value and risk – skills many people assembling these tools have never been trained in. Often, they don’t know what they don’t know.
In some organisations, thousands of citizen-developed agents have been developed. The 80:20 Pareto principle applies: a small proportion deliver real value, while a long tail consumes time, budget and elevates risk.
If we can’t stop this activity – and in most cases, we can’t – the alternative is to equip people properly. That means investing in skills and establishing governance pathways that are easy enough to use, and fast enough to keep pace with the pace of AI change.
Some organisations are already experimenting with this approach, using AI to help govern AI. Agents ask the right questions early – about data sources, decision impact or customer exposure – to guide people towards the appropriate risk pathway before problems emerge.
AI is changing the workforce and hybrid AI–human roles will become the norm, not the exception. If the CHRO doesn’t own the responsible AI conversation, who does?
Summary
Responsible AI is ultimately a people challenge rather than a technology challenge. As employees increasingly build, use and oversee AI tools, organisations can no longer rely on policies and controls alone. Effective governance depends on workers having the skills and judgement to understand AI outputs, assess risks and make informed decisions. With citizen development and hybrid AI-human work models becoming increasingly common, organisations must focus on capability-building, continuous learning and practical governance mechanisms, with CHROs ensuring the workforce is equipped to use AI responsibly, safely and effectively.
