Bloomberg News rolled out the use of artificial intelligence-generated summaries for its articles in January — but has had to add disclaimers for multiple articles that the summaries were in error.
The summaries “are evaluated by the company’s subject matter experts to continuously refine the large language model’s performance,” and help readers “save time through the efficient discovery of relevant and timely information, better decide which stories to read in depth, and enable them to explore and share more news,” the company said at the time.
A Google search reveals at least 20 instances of stories where summaries had to be removed after publication. For example, two summaries for articles this week on President Donald Trump’s tariff policy were taken down, one for a lack of attribution on when tariffs were due to kick in and another for unspecified “inaccuracy.”
Another article about the sale of steel mills had a summary taken down because the original incorrectly said that the United Steelworkers were opposed to the mill owner’s plans.
Bloomberg told the New York Times that these summary takedowns are by far the exception, with 99% of summaries meeting editorial standards.
“We’re transparent when stories are updated or corrected, and when A.I. has been used. Journalists have full control over whether a summary appears — both before and after publication — and can remove any that don’t meet our standards,” a Bloomberg spokesperson told the New York Times.
