David Koch is Compare the Market’s Economic Director.
Imagine an assistant that could carry the mental load of everyday life. It could book your appointments, order your groceries to be delivered, create an invite for your kid’s party, and even reach out to friends you’d lost contact with.
Probably sounds pretty good! We’ve all got so much to juggle. But would you give away your personal information and risk public embarrassment, or even financial loss, when things go awry?
Now picture a new acquaintance. Someone whose first name you remember but whose last name escapes you.
They’re pleasant, helpful and agree with you on most things. Over a few weeks you open up about the kids, the pets, work and the renos you’re planning. They even recommend a cheap tradie.
Then things get weird. They say they found your street address on an old post on your real estate agent’s Facebook page. The next week, they turn up on your doorstep unannounced. They apologise, then ask for access to your computer so they can start making bookings for you.
Do you hand over access? If the answer is no, think twice about Meta’s new AI assistant, Muse.
While the idea of a personal admin butler might sound appealing, some early users who’ve handed it the keys have been mortified by what it’s done with them.
Muse has just launched in Australia, after being downloaded millions of times in the US and Canada, where there are mounting reports of overreach.
Take tech writer Jason Aten, who says Muse synced his private messages even though he’d declined to give it access. Completely unprompted, it pitched him a column idea based on a private chat with his podcast co-host. Meta disputes his account, saying Muse can’t read messages unless users switch on two separate settings. Either way, it’s not a great look.
Last month, a TIME investigation found Muse updates dossiers on you and the people you mention in chats every hour. Basically, it’s got a running narrative on all your relationships – your shared interests, future plans and frictions.
And here’s the kicker. You don’t even have to use Muse to end up in it. If you message someone who does, and they’ve let it into that chat, your words can end up stored on Meta’s servers too.
Yikes.
And don’t get me wrong, I’m not a Luddite – in fact, I think AI can be really useful. I used it to create a charter for the grandkids outlawing things like jumping on the furniture and phones at the dinner table. Used this way, it’s fun and harmless.
What gives me the creeps is the potential for massive breaches of privacy when these seemingly helpful agents go rogue.
So how many of us are already letting AI near our money? Compare the Market asked more than 3,000 Australians whether they’ve used AI tools to help save time, earn money or make financial decisions.
Just over a third said they’d never used AI tools at all.
The most common use of AI was to improve productivity at work – cited by 22% of respondents – followed by researching products or services (21%).
But here’s the bit that worries me. 13% have already used AI to research financial decisions, 4% have asked it for advice on wealth management and 3% have used it for their tax returns. Those are exactly the people who’ll be tempted to hand an AI agent their bank card.
Have you used AI tools (such as ChatGPT, Copilot or Gemini) to help you save time, earn money or make financial decisions?
| To improve productivity at work | 22% |
| To help with study or education | 11% |
| To research products or services | 21% |
| To research financial decisions | 13% |
| To help manage a budget | 8% |
| To generate additional income or side-hustle opportunities | 4% |
| Created cheaper grocery lists | 6% |
| Investing | 4% |
| Help me with tax returns | 3% |
| Financial advice on wealth management | 4% |
| Mapping out household projects (renovations etc) | 4% |
| Support with DIY projects (helping me save money on tradies/service providers) | 6% |
| I have used AI, but not for any of these purposes | 15% |
| I have never used AI tools | 34% |
Source: Compare the Market survey of more than 3,000 Australians, August 2026.
Money is where it gets hairy. Under the ePayments Code, which most banks sign up to, you’re generally not on the hook for transactions you didn’t authorise. But if you gave a bot your card and told it to “always allow”, good luck arguing you didn’t authorise it.
Who’s responsible when an AI agent gets tricked, or simply blunders?
The grey area raises so many questions. Like, say a bot buys an insurance policy, would it even be valid? If your bot guesses the answer to a question and gets it wrong, people could be exposed to major financial losses if claims are denied. Or maybe the entire transaction is void, as an unlawful agreement between bot and business?
Even if these transactions are lawful, did the Act consider AI agents acting on behalf of consumers?
So, is the question now whether the law remains fit for purpose?
Current consumer safety nets and legislation aren’t ready for this kind of disruption.
And these tools are ripe for misuse by scammers too. Just eight days after Muse launched, a phishing campaign aimed at advertising managers started using a fake “Muse Ads” product as bait to steal their logins.
So, is Muse the best way to dip your toe into the world of AI agents? Personally, I’ll be avoiding this one – the risk/reward ratio doesn’t add up for me.
If you’re curious about Muse, start with the bare minimum. Don’t give it your passwords, don’t give it disk access, and only connect it to the apps you think it needs to deliver the support you’re looking for. And switch off the setting that lets Meta use your chats to train its AI.
You’ll also want to turn off any “always allow” switches – that basically gives it free rein to roam around your accounts as it sees fit. Toronto tech YouTuber Matt Robb learnt this the hard way. He asked Muse to sell a keyboard on Facebook Marketplace and ticked “always allow” during setup, assuming it would still check with him before finalising a sale. Instead, it accepted an offer, shared his home address and lined up a pickup. Next thing, a buyer turned up at his building.
But even if you’re being careful, can you really trust it? Can we really expect a product from a company that makes its money keeping us engaged to do only what we say?
Meta claims it keeps your passwords and payment details out of the agent’s view and only looks at messages when allowed. To be fair, it’s put real work into security, and it’s offering up to $300,000 to anyone who finds a hole. But as we’ve seen, the permissions are confusing enough to catch out seasoned tech reviewers.
Unlike that freaky acquaintance who keeps showing up, you can’t just call the police before things escalate.
The Government is rightly moving quickly to make social media and AI services safer online. The same urgency is now needed to protect Australians if these agents start spending our money.
There’s no time for musing. Muse is here now, so let’s fix this before it’s too late.
David Koch is Compare the Market’s Economic Director.



