LA Times uses AI to analyze articles for political bias

Imagine writing a deeply reported investigative piece on the historical legacy of the Ku Klux Klan in local politics, only to find your own paper's AI tool slapping a bullet point at the bottom suggesting the hate group was simply a response to social changes.

That is precisely what just happened at the Los Angeles Times.

Yesterday morning, LA Times billionaire owner Patrick Soon-Shiong published a letter to readers announcing that the paper is deploying AI to assess stories for political stance and bias. Articles deemed to take a perspective will now receive a "Voices" tag, along with AI-generated "Insights" at the bottom offering "Different views on the topic."

An immediate editorial nightmare

Soon-Shiong explained that "Voices" will not just cover traditional opinion pieces, but also news commentary, analysis, and reviews. In his view, presenting different viewpoints fulfills the outlet's journalistic mission.

Editorial staff, however, are furious. LA Times Guild vice chair Matt Hamilton told The Hollywood Reporter that while the union supports helping readers differentiate news reporting from opinion, unvetted AI analysis will do very little to improve trust in media.

Within 24 hours, the tool produced glaring embarrassments.

According to The Guardian, a March 1st op-ed discussing the dangers of unregulated AI in historical documentaries was tagged as "Center Left," with the AI helpfully offering a counter-argument that AI actually democratizes storytelling.

Even worse, an investigative piece about California cities electing KKK members in the 1920s received a bullet point minimizing the group's threat as a mere artifact of white Protestant culture. The author noted on X that while that reflects historical framing from that era, the AI clumsily surfaced it as a counterpoint to modern reporting on neo-Nazi legacies. The paper quietly deleted that insight shortly after.

Other newsrooms use AI—including The New York Times, The Washington Post, and Bloomberg—but generally for back-end research or administrative tasks, not automated editorial commentary without human oversight.

Why it matters

For regular readers, this feature adds noise rather than balance. Instead of finding helpful historical context, subscribers are greeted with tone-deaf automated rebuttals that feel glued onto the end of serious investigative reporting.

For the media industry, the rollout serves as a cautionary tale against outsourcing editorial judgment to generative models. Publishers are desperate to shed allegations of partisan bias, but using synthetic text to manufacture false balance on settled issues like hate groups is dangerous.

We think Soon-Shiong's intent to bridge political divides might look sensible on a corporate slide deck, but bypassing human copy editors was an unforced error. Journalism requires moral clarity and contextual nuance—two things language models simply cannot supply when trying to play devil's advocate.