Michelle ThompsonApple has temporarily disabled its newly launched AI feature, which summarizes news notifications, after it repeatedly delivered error-ridden headlines, causing backlash from a news outlet and press freedom organizations.
This rare reversal by Apple comes after the AI generated misleading or entirely false news summaries that closely resembled standard push notifications.
On Thursday, Apple released a beta update for developers that deactivated the AI news and entertainment summary feature. The company intends to refine the AI before re-enabling it in a future update. Additionally, the update will make it clearer to users that the summaries are AI-generated, signaling the possibility of inaccuracies.
The BBC raised concerns about the feature last month, requesting its removal after it wrongly reported that Luigi Mangione, charged with the murder of the UnitedHealthcare CEO, had shot himself. Another incident saw a notification merging three New York Times articles, falsely claiming that Israeli Prime Minister Benjamin Netanyahu had been arrested.
In December, a BBC spokesperson told CNN that it was crucial for Apple to address these issues swiftly, as accurate news is vital for maintaining trust. The spokesperson stated that the AI summaries did not reflect the original BBC content, and in some cases, contradicted it entirely.
On Wednesday, the AI feature once again misrepresented a Washington Post notification, incorrectly stating, “Pete Hegseth fired; Trump tariffs impact inflation; Pam Bondi and Marco Rubio confirmed.” These statements were false.
Geoffrey Fowler, a tech columnist for the Washington Post, criticized the feature, calling it “wildly irresponsible” and suggesting that Apple should disable summaries for news apps until it improves the AI’s accuracy.
Press freedom groups have also raised alarms about the potential harm the summaries could cause, with Reporters Without Borders calling them “a danger to the public’s right to reliable information on current affairs.” The National Union of Journalists urged that the AI summaries be removed, emphasizing the need for the public to trust the accuracy of the news they receive.
Apple’s AI issues reflect broader concerns with AI technology, which is known to occasionally generate “hallucinations,” or fabricated information. As Suresh Venkatasubramanian, a Brown University professor, explained, large-language models like ChatGPT are designed to provide plausible-sounding responses, but these may not always be factual.
A study in July 2024 from Cornell, the University of Washington, and the University of Waterloo found that AI models still can’t be fully trusted, due to their tendency to create inaccurate information.
