It’s no surprise organisations are moving fast on AI, with 88% of them using it in at least one core business function. But the perils of AI could be lying dormant and ready to erupt inside the very people that AI is meant to be enabling: our employees.  Related Article Block Placeholder Article ID: 340252 Australia can’t hit its 1.2 million tech worker target if experienced women keep leaving Tegan Jones In March, the Harvard Business Review reported on a study by the Boston Consulting Group into a condition called AI “brain fry”, which workers have described as a buzzing sensation, mental fog, difficulty focusing, and slower decision-making caused by intensive oversight of AI agents and tools. Importantly, AI brain fry is not burnout. Burnout builds slowly through emotional and physical depletion over months, whereas AI brain fry is fast and specific: with working memory, attention systems, and executive control pushed past their limits, often within a single session.  AI brain fry is said to be causing an increase of 40% in major workplace errors. Yes, major errors, not just typos. Even more concerning, the study found workers experiencing brain fry were 39% more likely to say they intend to quit than those who weren’t. Separately, Upwork Research Institute research found managers expect employees to take on a wider range of responsibilities, handle more complex projects, and maintain continuous productivity in return for AI’s contribution to lower-order tasks. It’s fair to say these employees see it differently, because that same study found 77% of them feel AI adds to their workload. What’s the cure for AI brain fry? The key to unlocking a cure for AI brain fry has very little to do with AI and everything to do with how organisations design work around it. Smarter business news. Straight to your inbox. For startup founders, small businesses and leaders. Build sharper instincts and better strategy by learning from Australia’s smartest business minds. Sign up for free. * indicates required Email Address * By continuing, you agree to our Terms & Conditions and Privacy Policy. The solution doesn’t sit in your tech stack but in understanding and involving employees in the effects of AI in the workplace. Simply adding AI on top of existing demands, without redesigning the work around it, is the very flaw producing all the harm.  The International Labour Organization’s April 2026 report about the effects of AI on the psychosocial work environment argues that psychosocial risks arising from the use of AI require an integrated regulatory response, treating worker health as a first-order governance concern alongside data protection and bias auditing. And Australian regulators agree. The WHS Amendment (Digital Work Systems) Act 2026, passed in New South Wales in February, is the first Australian legislation to name algorithms, AI, and automated work allocation as explicit sources of WHS risk. Under this law, if a digital work system creates psychological harm – through excessive demands, unreasonable performance metrics, or intrusive monitoring – the organisation deploying it is on notice that it will be legally responsible.  Related Article Block Placeholder Article ID: 331126 The debate over NSW workplace AI oversight laws, explained David Adams Vendor-built does not mean off the hook. The research consistently shows the most effective control for AI-driven harm is not technical, but organisational. And it starts with a conversation. Unless Australian organisations start to tackle this issue head-on from an occupational health perspective, it could be more than just the brains of heavy users of AI at work heading for the frying pan. The top five actions all Australian organisations must take today: Map your digital work systems: Every algorithm, AI tool, or automated platform used to allocate work, set targets, or monitor performance is a starting point for risk assessment under the Act in NSW. Consult workers before and after deployment: Under WHS laws, consultation is mandatory at every stage of the risk management cycle. Ask directly: what does this tool change about your day? What control have you lost? What pace does it set? Those answers are your hazard identification data. Audit the work design, not just the system: The hazard is rarely the technology. It is whether the way you have deployed it – the demands it creates, the autonomy it removes, the transparency it withholds – produces conditions that cause harm. Build human checkpoints into automated decision-making: The law requires human oversight of algorithmic work allocation. If consequential decisions are being made without a review point, that is a design gap with legal exposure. Review your controls on a set cycle: Controls that are not reviewed do not count as controls. The organisations ahead of this are not waiting for a regulator to draw the line. They are treating AI governance and psychosocial safety as the same conversation – because under Australian law, they now are. Stay in the know Never miss a story: sign up to SmartCompany’s free daily newsletter and find our best stories on LinkedIn.