Apple is temporarily pausing its artificial intelligence software’s ability to summarise notifications from news and entertainment apps, after concerns were raised over inaccuracies.
Apple Intelligence’s notification summaries feature was designed to “help users prioritise and stay in the moment” by using AI technology to summarise large numbers of notifications for things such as messages, emails, and news alerts, the company said.
But the feature has been criticised for sometimes inaccurately summarising news headlines and even treating scam and spam messages and emails as legitimate.
The BBC complained to Apple about the feature in December, after Apple Intelligence generated an incorrect headline about Luigi Mangione, the former tech worker accused of shooting dead a healthcare company CEO in New York City, which falsely claimed Mangione had shot himself.
The generative AI feature uses a Large Language Model (LLM) which statistically predicts each word it should use — a technology which is yet to be perfected and has a tendency to sometimes hallucinate incorrect information.
Apple Intelligence’s notification summaries have also allegedly hallucinated incorrect sport results, and even falsely stated Israeli Prime Minister Benjamin Netanyahu had been arrested.
Image: Bluesky
Apple says AI ‘will occasionally make mistakes’
Last week, with the release of its latest beta software to developers (iOS 18.3, iPadOS 18.3, and macOS Sequoia 15.3), Apple stopped allowing notification summaries to be generated for news and entertainment apps.
It began warning users that the feature was in beta and “will occasionally make mistakes that could misrepresent the meaning of the original notification”.
The company also began to display AI-generated notification summaries in italicised text, in an attempt to distinguish them from regular notifications.
Notification summaries for other apps such as communication and social apps handling messages and emails have remained available.
It is unclear when iOS 18.3 will be released publicly, but it is expected in the coming weeks.
Apple Intelligence was publicly released in the US in October 2024 and made available in Australia in December. Image: Apple / Supplied
Apple did not respond to a request for comment, but previously told US media it was working on updates to the notification summaries feature.
“A software update in the coming weeks will further clarify when the text being displayed is summarisation provided by Apple Intelligence,” the company said.
“We encourage users to report a concern if they view an unexpected notification summary.”
AI notification summaries can also be turned off completely or enabled only for specific apps on a user’s device.
The UK’s National Union of Journalists, which raised concerns over “the erosion of public confidence in journalism” if the feature was not removed, said it would now push for “increased scrutiny addressing inaccuracies as part of any developments to the feature”.
And an example of how a recent @Telegraph notification was summarised by Apple Intelligence too pic.twitter.com/bMHI387O7e
— Nick Sutton (@suttonnick) January 15, 2025
Concerns over scam and spam messages
Apple users have also raised concerns after their devices’ Apple Intelligence systems mistook obvious scam emails and spam messages for legitimate correspondence which required user action.
An Australian user reportedly told Crikey his iPhone had prioritised and summarised a scam email which purported to be from the Australian Taxation Office.
“Income Tax 751.23 AUD for the period Nov – Dec 2024 is pending preparation for Lodgement,” the notification reportedly said.
The new @Apple intelligence feature flagged an email as priority, but really it’s a scam email about a security breach from Apple. Lol talk about ironic. Needs more intelligence I suppose pic.twitter.com/iwMhTXi0Du
— yeaimightpre (@YeaiMightPre) January 15, 2025
Apple Intelligence launched in Australia and New Zealand in December 2024, after being debuted in June of that year and rolled out in the US in October.
The features are available on iPhone 15 Pro and newer, Macs with Apple’s M-series chips, and iPads with an A17 Pro chip, M1 chip, or newer.
Several other genAI systems have also caused public concern after hallucinating inaccurate or dangerous content.
Google’s Gemini AI has famously told users to eat glue and rocks, and even sent a student an unprompted death threat.
Apple was relatively late to add generative AI features to its devices, compared with competitors such as Google and Samsung.
The tech giant also rarely rolls back new features, and often prefers not to rush implementation of new tech before it is perfected.
Peter Sobot, a software engineer at Spotify, wrote on X that he could no longer trust Apple Intelligence’s notification summaries.
“I know how LLMs work, and I know this is a very hard problem (especially on device!), but I’m still shocked Apple shipped something less reliable than 99.9999%,” he said.
- This story first appeared on Information Age. You can read the original here.
