Apple’s new artificial intelligence features are rephrasing scam messages and emails to iPhone and Mac users to make them look more legitimate and then flagging them as a priority, raising concerns that it will lead to more people falling for them.
Late last year, Apple rolled out its AI-powered “Apple Intelligence” update to millions of Australian iPhone, iPad and Mac computer users.
Its flagship features include notification summarising and prioritisation abilities. These use AI technology to summarise the meaning of multiple push notifications in a single message, or to flag certain alerts as a priority.
Apple’s example of this summarising in action shows a group chat with friends that has been condensed down to a single notification that says: “Brunch after soccer on Saturday; restaurant or host at home suggested.”
Similarly, Apple’s prioritising feature example, which the company promotes as flagging the “most urgent emails”, shows it highlighting emails from friends, colleagues and an airline about imminent demands.
Apple has already come under fire after these features repeatedly incorrectly summarised BBC headlines, such as by falsely claiming that UnitedHealthcare CEO Brian Thompson’s alleged killer Luigi Mangione had shot himself.
This issue extends to people’s personal communications, too. Screenshots from iPhone users show this new suite of AI-powered features appears unable to distinguish between messages sent by real individuals and organisations and fake requests made by scammers imitating others. In fact, the AI-powered features may even make it harder for users to initially distinguish between real and fake messages.
Steve, a pseudonym granted as his work has not authorised him to speak to the media, was surprised to see that his recently updated iPhone had prioritised and summarised an email saying that he had to lodge a income statement to the Australian Tax Office.
“Income Tax 751.23 AUD for the period Nov – Dec 2024 is pending preparation for Lodgement,” said the highlighted notification.
This was nonsense. Steve was disappointed that Apple’s new features had flagged this obvious scam message as a priority.
“It’s fun to make fun of AI, but having an AI tool that’s onboard millions of devices enable scams like this is going to catch people out and cost them money,” he said in a message.
Other people have posted similar examples on social media. UQ honorary professor Jeremy Howard shared a screenshot on Bluesky of a prioritised alert, remarking “Maybe Apple Intelligence shouldn’t mark scam emails as ‘Priority’ with a summary saying it’s for security purposes?”
Another Bluesky user, @colin.hoagie.fun, shared a screenshot of an Apple-summarised text message purporting to be from American postal service USPS, suggesting “Imagine if Apple Intelligence actually flagged the scam messages instead of helpfully summarising them?”
Daswin De Silva, La Trobe University’s full professor of AI and analytics and deputy director of the Centre for Data Analytics and Cognition, says Australians, who reported losing $2.7 billion to scams in 2023, will be at a greater risk because people will trust Apple’s summaries.
“Because of the AI hype, lots of experts saying this is really good, the government funding lots of AI programs, we hear there is AI in my phone that we can use to be more efficient or be more productive, then we will trust this AI,” he said.
De Silva said summaries might make people more likely to fall for scams by removing indicators of legitimacy so that all messages, scam or not, look the same. “By trying to reduce the information, you’re actually creating more work for the end user or the consumer to evaluate the information and then look for explainability factors and go into the actual messages and start reading them all over again.”
De Silva said companies have been impatient with releasing new AI features, dealing with any issues after they were released.
“People are still getting used to working with AI or living with AI, and that’s why it has to be released in a gradual process. Dropping new features every so many months really doesn’t help.”
Apple did not respond to a request for comment, however the company told the BBC earlier this week that it would update the feature to “clarify when the text being displayed is a summarisation provided by Apple Intelligence”.
Would you trust a text or email summary provided by AI? Write to us at letters@crikey.com.au. Please include your full name to be considered for publication in Crikey’s Your Say. We reserve the right to edit for length and clarity.