As businesses across Australia explore the benefits of generative AI in the workplace – from increased productivity, to better employee experiences – millions of workers may be embracing the tools without fully realising the potential risks.

A survey Google conducted with IPSOS in January found that almost half of all Australians use generative AI and almost 75 per cent of those report using it for work.

Meanwhile, a survey conducted by HR platform Workday revealed that about 65 per cent of Australian workers confirmed their employer had introduced AI in the workplace.

But even using generative AI for a task as simple as sending a business email can have unintended consequences.

End Of Lease Cleaning Melbourne director Michael learned the hard way when a mistake in a seemingly harmless business email cost him $2000.

It used to take his team about five hours to respond to customer emails so they started using a generative AI tool to speed up the process.

”We were trying to save some time by not typing individual lists of cleaning services,” he told 9news.com.au.

Instead, they would input a prompt outlining the services a customer required and have the AI tool generate an email detailing the services, their costs, and a job quote.

The tool slashed their response time down to one hour, but Michael admitted it wasn’t perfect.

On several occasions, the AI tool mistakenly listed a ‘full wall clean’ instead of a ‘spot wall clean’ but did not change the quote to reflect the more expensive service.

It meant Michael and his team had to provide the $500 to $700 full wall clean at the much lower price of the spot clean, losing the business hundreds.

The final straw came in March, when Michael had the AI tool generate a quote for a “filthy” property that required about $2000 worth of cleaning.

He gave the AI generated email a quick once-over then sent it to the customer.

Michael didn’t realise it was riddled with errors until a week later, by which time the customer had gone to a different company.

“We lost quite a lot of money,” Michael said.

He’s not the only Australian worker paying for mistakes made by generative AI in business emails.

Others who spoke to 9News claimed that AI had addressed customers, clients and colleagues by the wrong name or title in emails, jeopardising business opportunities and working relationships.

But the risks go beyond awkward mishaps, Dentons Intellectual Property and Information Technology lawyers Robyn Chatwood and Michael Park told 9news.

Generative AI tools will “hallucinate”, wherein they make up facts that aren’t accurate, which can cause more problems for workers if they include these “hallucinations” in professional correspondence.

It’s also not uncommon for AI tools to infringe on copyright or mistakenly breach confidentiality rules, which can have serious ramifications in a professional setting.

In such situations, workers “still have the responsibility and the liability” according to Chatwood.

“You can’t just say the machine made a mistake, because you should have checked it,” she said.

Park warned that the best way for Australian workers to protect themselves from these kinds of mistakes is to stick to their employer’s AI use policy, no matter how tempting it may be to speed up a task by using AI.

“If your policy says don’t do it, then just don’t do it,” he told 9news.

“You’re protecting yourself from potentially getting into trouble.”

Workers or small business owners who don’t have an AI use policy should err on the side of caution, he added.

Though it means their response time is back at the five-hour mark, that’s better than making another costly mistake using generative AI.

“If you are using AI, you definitely need to read everything two to three times before you send that email,” he said.