By Liam Reilly | CNN
The Los Angeles Times’ billionaire owner, who unveiled an AI tool that generates opposing perspectives to be displayed on opinion stories, was unaware the new tool had created pro-KKK arguments less than 24 hours after it launched — and hours after the AI comments had been taken down. The incident presents a massive hurdle for the Times, which looks to win back old subscribers and woo new ones with a new suite of offerings.
During an interview with CNN on Tuesday, Patrick Soon-Shiong, the Times’ executive chairman, admitted he had seen neither the piece nor the AI response. But he said the content’s removal showed that there are operational “checks and balances” to the recently introduced system, pegging the moment as a learning opportunity.
“(The incident is) a good lesson to show that at least artificial intelligence is not fully there yet (…) it’s in an attempt to understand that,” Soon-Shiong said.
On early Tuesday, the new AI tool generated counterpoints to a February 25 column from Times columnist Gustavo Arellano. Arellano’s column argued that Anaheim, California, ought to not forget the Ku Klux Klan’s role in its past — calling the white supremacist group “a stain on a place that likes to celebrate the positive” — and connecting it to today’s political landscape. But the divergent views generated by the Times’ AI produced a softer vision of the far-right group, which it called “‘white Protestant culture’ responding to societal changes rather than an explicitly hate-driven movement.”
While the AI-generated comments have since been removed from the piece, and Arellano chimed in to say the “AI actually got that right” since “OCers have minimized the 1920s Klan as basically anti-racists since it happened,” the newspaper owner’s lack of awareness about the controversy is a glaring issue.
Arellano’s piece is not the only one to have contained an AI-generated error or misleading comments within 24 hours of the AI tool’s introduction. An op-ed from Scott Jennings about President Donald Trump’s response to the Los Angeles wildfires came under scrutiny after the Times’ tool labeled the piece as centrist despite its right-leaning talking points. The story’s AI-generated counterperspective also failed to note that Trump had threatened to withhold federal aid to Los Angeles unless its leaders complied with specific demands.
The AI tool, which is dubbed “Insights,” is part of a suite of announcements unveiled by Soon-Shiong on Monday. In addition to the Insights button that appears alongside all opinion stories — a section that has now been rebranded as Voices — Soon-Shiong also debuted a bias meter and live-stream programming from LA Times Studios.
The changes come several months after the Soon-Shiong announced his intention to add AI tools to the website, part of the publisher’s push to appeal to younger audiences and conservatives. They also come after months of turmoil have rocked the outlet, including an exodus of readers and editorial board members following a blocked op-ed, mass layoffs, voluntary buyouts, a botched edit that reversed a writer’s position and comments from Soon-Shiong that undermined his own reporter.
Voices and Insights
Despite the immediate alarming results, Soon-Shiong told CNN that Insights’ AI-generated pieces do not seek to create divisive responses but, rather, inclusive ones.
“It’s actually trying to say, ‘OK, this is what this piece is trying to say with (…) all the references from,’” Soon-Shiong said. “And then, if you don’t agree with what this piece is saying — because it’s a voice, it’s not news — this is the opportunity for us to share with you an alternative view that somebody else would look at tied to its references.”
If readers disagree with a Voices piece, the Insights button should provide an alternate view, he said. He added that the new Voices hub stems from that inclusive inclination and a desire to clearly separate the opinion section from the newsroom.
Insights deploys the Times’ in-house Graphene AI content management system to determine a story’s bias, running a word-by-word analysis to generate an alternate view. Graphene was trained by partnering with external AI models and using decades of Times content and historical pieces, all of which were validated using machine learning and an editorial review process to establish a bias scale, Soon-Shiong said. In his Monday letter to readers, Soon-Shiong described the tool as “an experimental, evolving technology.” But he told CNN that a team spent months continuing to develop the tool after he initially teased it in December. Like other AI platforms, the tool will continue to learn as it ingests more Voices.
AI-generated perspectives are accompanied by sourcing so that readers can learn more about a particular position. But a blistering Nieman Lab review found several problems with the sources and even the way they’re cited. On some occasions, the AI tool cited mediocre or less-than-reputable sources, while on other occasions it duplicated sources in citations. Elsewhere, the AI tool used citations to misleading effect, articulating a point missing in the cited source.
To correct errors, Soon-Shiong says there is some level of human oversight but it’s hard for the team to validate all responses in real-time given the scale at which the tool is being applied. Beyond the team in charge of overseeing the tool, readers who encounter AI errors are encouraged to report issues on the Times’ Insights page or in the comments section.
The Times is hardly the first publisher to imbue its website with AI tools, and most newsrooms or news tools that have added AI to writing or reporting have yielded disastrous results. Two years after OpenAI launched ChatGPT, even the most sophisticated AI chatbots remain plagued by accuracy and reliability issues.
There are, of course, streamlining benefits to AI in the newsroom: AI tools can expedite newsgathering and, with human oversight, catch problems in articles. But publishing unvetted content penned by AI is a massive gambit — it risks delivering inaccuracies that misinform readers and undermine trust.
LA Times Studio and LA Times Next
The live-stream element — a project dubbed “The Stream” from LA Times Studio, the Times’ production studio that’s behind the outlet’s TV, film and audio projects — was also introduced Monday. Under the venture, the studio will produce 12 hours of original live-streaming per day spanning news, entertainment, food, business, culture, lifestyle and true crime. To fuel the massive undertaking, LA Times Studio is working closely with the newsroom, Soon-Shiong said, but it’s also onboarding studio engagement staff to handle the technical aspects of streaming, shooting and producing.
Absent from Soon-Shiong’s list of announcements on Monday was any mention of LA Times Next, a project the billionaire has been quietly developing alongside the others.
In February, Status described LA Times Next as “a new entity that will prominently feature digital-first personalities, many of whom will appeal to the MAGA base.” But Soon-Shiong dismissed that description as “speculation.” While he declined to clarify what the project will entail, he told CNN that LA Times Next will be located in Los Angeles, Washington, DC, Nashville and several other places nationwide.
Soon-Shiong also declined to comment on Status’ report that Jennings is attached to the project and that there have been conversations about adding Candace Owens and Ana Kasparian.
“We do things in a very thoughtful way to involve the community and, when we release LA Times Next, we will then ‘inform the truth’ rather than have so-called journalists write speculation and then ask us to respond to speculation,” Soon-Shiong said.
Times readers can expect to hear more about LA Times Next in three to six months, he said.
The editorial board
The billionaire’s other project for the Times is rebuilding the paper’s editorial board, an especially urgent task given Carla Hall, the Times’ last remaining editorial board member, accepted a buyout last week after 32 years with the company.
Hall’s departure follows her peers’ exodus after Soon-Shiong blocked an op-ed endorsing then-Vice President Kamala Harris for president at the 11th hour in October. The move has been publicly critiqued by Times readers and staffers alike, many of whom ended their subscriptions or left the paper over allegations that the Times owner preemptively bent the knee to Trump.
The Times owner has previously said that the old editorial board “veered very left” and that a new board should “have somebody who would trend right, and more importantly, somebody that would trend in the middle.” While the billionaire still won’t say who will staff a new editorial board, he said that “it’s time for a rebirth and reinvigoration of a next generation of editorial members who recognize the new world as we see it.”
The changes are an attempt to regain old subscribers while courting new ones at a difficult time for the news industry. Like Soon-Shiong’s Times, Jeff Bezos’ Washington Post has lost subscribers over his involvement, hemorrhaging 250,000 subscribers in October after a pro-Harris endorsement was blocked and at least 75,000 in 48 hours last week over his opinion section overhaul. The Times lost more than 7,000 readers in October, almost 2% of its total subscription base.
Bezos’ Wednesday shakeup of the Post’s opinion section has similarly left staffers in a state of rebellion, possibly worsening the low morale that has for months precipitated several staff resignations and departures at the Post. But Soon-Shiong, whose last few months have resembled the billionaire Post owner’s, applauded the move on X: “Welcome to the club Jeff!”
Already, Soon-Shiong claims he’s received calls from old subscribers looking to resubscribe.
“It’s only one day, but, again, I think engaging all Americans, and the opportunity to have engagement with the young — the younger readers — and to create a platform that allows interactivity, connectivity is the best that we can try,” the billionaire said.
But what a day it was.
The-CNN-Wire
& © 2025 Cable News Network, Inc., a Warner Bros. Discovery Company. All rights reserved.