Australians have been urged to think twice before using Meta’s Muse AI assistant, which can perform tasks on a user’s behalf but has raised concerns over privacy, security and financial responsibility.
Muse went live in Australia on Wednesday and is designed to connect with services people already use. It can make calls, search emails, order groceries and book restaurant reservations, with a digital teddy bear-like character called Jolly acting as its public face.
The assistant has attracted 10 million users worldwide since launching in September. Its capabilities extend beyond answering questions, allowing it to research purchases and take action for users.
Tech Guide editor Stephen Fenech said people were accustomed to using AI to ask questions, but Muse offered something different.
“It can actually do things for you, find things out for you, act on your behalf,” he said.
Meta spokesperson Jason Juma-Ross said users could tell Muse which gifts they needed to buy for family members and ask it to carry out the research.
The tool can be used to buy make-up from MECCA, stationery from Officeworks and tickets through Ticketek. It can also search property listings on Domain and Realestate.com.au.
Concerns over Meta’s Muse AI assistant
Questions have been raised about the way the assistant handles personal data. Reports have alleged that Muse read messages it was not meant to access, suggested ideas based on private conversations and disclosed a user’s home address while helping to sell a keyboard.
It has also been reported that the system can build a background profile on people who do not use it, based on conversations with those who do.
David Parry, from Murdoch University’s School of IT, said AI agents could “overreach”, be mistaken for malicious bots by businesses receiving bookings and potentially be exploited by scammers.
“It’s version one. There are definitely people who are going to be finding this sort of thing useful,” he said.
Mr Parry also questioned whether users could rely on such services remaining available in the long term, saying he would be more comfortable dealing with a paid provider.
“Anything free is always a bit suspicious,” he said.
Financial risks have prompted a further warning, with research from Compare the Market finding that just over one third of those surveyed had never used AI tools. The research found that 13 per cent had asked AI for financial advice, 4 per cent had sought wealth management tips and 3 per cent had used it to complete tax returns.
Compare the Market economic director David Koch said those users could be tempted to give an AI agent access to their bank card.
“Money is where it gets hairy,” he said, warning that people could face disputes over whether a transaction had been authorised if they had instructed a bot to allow payments.
He also questioned who would be responsible if an agent were tricked or made a mistake, including whether an insurance policy bought by a bot would be valid or whether a transaction could be deemed void.
Meta says users remain in control of what they share and that passwords and payment methods are kept out of the agent’s view. Mr Juma-Ross said access could be revoked at any time.
The company is offering a bug bounty of up to 300,000 dollars to anyone who identifies a flaw in the system.
