Apple’s plan to streamline news notifications backfired spectacularly in December. The trouble began with the BBC. The corporation filed a complaint after spotting a screen capture of a BBC News iPhone notification being widely shared online that implied the BBC had made an error. It stated Luigi Mangione – arrested over the murder of a healthcare executive – had killed himself. He hadn’t. But the BBC hadn’t actually said this either.
This wasn’t a one-off. Next came screen grabs of further notifications that claimed tennis superstar Rafael Nadal had come out as gay. He hadn’t. Another stated Luke Littler had won a darts championship. The competition hadn’t even started. In one standout case, a Washington Post notification managed to get every single fact wrong (see below). What was going on?
This is my periodic rant that Apple Intelligence is so bad that today it got every fact wrong its AI a summary of @washingtonpost.com news alerts.
It’s wildly irresponsible that Apple doesn’t turn off summaries for news apps until it gets a bit better at this AI thing.
— Geoffrey A. Fowler (@geoffreyfowler.bsky.social) January 15, 2025 at 6:15 PM
The issue wasn’t the news outlets themselves. A small arrow icon in each notification betrayed them as being summaries generated by Apple Intelligence. However, you might argue making summaries condensed from actual headlines that mangled facts doesn’t seem very intelligent. And the incident couldn’t have come at a worse time, with trust in traditional news and media rapidly eroding.
Apple’s intent had been to save people time. Notifications can be a distraction and overwhelming. Previous iOS updates made it possible to group notifications and lowered visual load. Apple Intelligence reduced the number of visible notifications further – but at the cost of accuracy. And yet these notifications retained each news app’s icon, making it appear that the news source – rather than Apple’s tech – was responsible for their content.
The challenge for Apple is that large language models (LLMs), like the one at the heart of Apple Intelligence, lack true intelligence. They don’t understand facts and context. They are closer in nature to fancy autocomplete, like when you build a sentence by repeatedly tapping suggested words on your iPhone’s keyboard. Only LLMs are trained on gargantuan data sets, increasing the probability of responding to prompts (in this case, summarizing news headlines) with a result that’s at least coherent.
However, with few words to work with, these systems struggle to be factually correct. Even if the summaries had been generated from the full news articles, it’s likely Apple Intelligence would have struggled to condense several – often unrelated – news pieces into the tiny space afforded to a notification. In short, with LLMs being based around probability rather than accuracy, errors are inevitable. And because these systems are automated, there is no human oversight to catch mistakes before users see them.
Apple’s eventual response was tepid, mostly noting Apple Intelligence was still in beta. The company promised changes that would make it more obvious (as in, at all obvious) summaries may contain errors. Since then, iOS 18.3 has temporarily disabled all summaries from news and entertainment apps, and italicized those from other apps to differentiate them from standard notifications.
Apple has heavily marketed Apple Intelligence since September 2024.
Those changes are welcome. The excuses are not. Labeling something ‘beta’ doesn’t absolve a company of responsibility, especially in this case when Apple has been heavily promoting Apple Intelligence as the flagship feature of its iPhones since September 2024. Any suggestions of shortcomings were, at most, relegated to footnotes. The reality is these summaries should never have been released in such error-prone form, leaving users exposed to a firehose of misinformation.
However, Apple is not alone. AI is rapidly infiltrating various technologies, and similar ‘hallucinations’ (the industry term for AI fabrication) are popping up everywhere, from email inboxes to search engines. Notably, Google – once a fast, efficient means to access a wealth of information online – now places an AI summary before search results. In a recent case, its AI insisted water below 32°F wouldn’t freeze, because it had to reach freezing point first. This followed another gem where Google’s AI, not understanding context, lifted a suggestion from a satirical Reddit post that recommended using non-toxic glue to make cheese stick better to pizza.
In isolation, that might have been funny. But none of this is amusing during a time when we are bombarded with disinformation, making it increasingly tricky to tell fact from fiction. There’s enough fake news out there already; we don’t need AI generating yet more of its own.
To disable notification summaries on iPhone, open Settings and head to Notifications > Summarize Notifications. They can be turned off entirely or on a per-app basis.