GettyImages 2193585826

The biggest story in journalism right now is that CBS News agreed to give Donald Trump $16 million in a legally blessed bribe. The great sin of “The House That Edward R. Morrow Built” involved 60 Minutes airing a run-of-the-mill interview with Kamala Harris that made her look like a competent public servant with years of experience. Since Trump’s interviews, regardless of editing, sound like a dementia patient navigating a law school cold call, he decided CBS had committed consumer fraud because Harris spoke in complete sentences.

But apparently we weren’t done with today’s “dystopian assault on freedom of the press” news! And it came after an unlikely target: Law360. I certainly didn’t have “legal industry trade publication” on my censorship BINGO card. Then again, Biglaw lateral moves have suddenly become political stories so perhaps this marks inevitable cowardice creep reaching the legal press.

But the part of this story that elevates it from ominous development for civil liberties to comi-tragic is that Law360 is owned by LexisNexis and therefore the agent of Law360’s doom is… an AI algorithm! A new bias detecting ChatGPT wrapper slapped together by some LexisNexis product engineers probably taken away from actually useful work to build a degenerative AI to strip news articles of any semblance of value. 2025, man… Does. Not. Miss.

NiemanLab, Harvard’s digital journalism center, reports that Law360 has ordered its reporters run their stories through an AI bias detector designed for “applying a neutral voice to copy” and to be mandatory for “headline drafting, story tagging, and ‘article refinement and editing.’”

As one might imagine the journalists, represented by the Law360 union, object to this half-baked idea. A policy this ethically bankrupt could only arise from non-journalist executive input.

The announcement came a few weeks after an executive at Law360’s parent company accused the newsroom of liberal political bias in its coverage of the Trump administration. At an April town hall meeting, Teresa Harmon, vice president of legal news at LexisNexis, cited unspecified reader complaints as evidence of editorial bias.

Giving uncritical weight to squeaky wheel complaints, especially in an environment where a government official weaponized his followers to act on their every grievance up to and including STORMING THE FUCKING CAPITOL, is a dunderheaded management strategy only an MBA could come up with. But it’s almost certainly a cynical one. If we all start writing complaints that the headlines are neutered doublespeak, will Law360 be ordered to reverse course? I’m incredulous.

While the article notes that there’s not an established throughline from those remarks to the implementation of the policy, it speaks to a mindset that clearly got out of hand.

But let’s put aside the wisdom of the policy and focus on the fact that the bias detector is also terrible at its job. Because that’s just a little bit more fun. Only at a tech company could someone think that generative AI tools being developed for dedicated legal work tasks could be bolted onto the editorial process of a news publication.

Generative AI is a powerful tool in the same way a screwdriver is a powerful tool. But you wouldn’t use a screwdriver to do your taxes. Yet that’s the thinking involved in bringing AI into an editorial process. To borrow from the TV series Veep, it’s like using a croissant as a dildo: “It doesn’t do the job, and it makes a fucking MESS!”

She also criticized the headline of a March 28 story — “DOGE officials arrive at SEC with unclear agenda” — as an example. In the same town hall, Harmon suggested that the still experimental bias indicator might be an effective solution to this problem, according to two employees in attendance.

But… DOGE officials did arrive at the SEC with an unclear agenda. The White House couldn’t be clear about who was running DOGE let alone its agenda. This is just a factual statement that, if anything, is biased in favor of DOGE since its suspected agenda to steal data and hamper regulation was about as disguised as three raccoons in a trench coat.

The report notes another story about the Trump decision to mobilize the California National Guard:

Several sentences in the story were flagged as biased, including this one: “It’s the first time in 60 years that a president has mobilized a state’s National Guard without receiving a request to do so from the state’s governor.” According to the bias indicator, this sentence is “framing the action as unprecedented in a way that might subtly critique the administration.” It was best to give more context to “balance the tone.”

It was the first time in 60 years though! That is the relevant context. As is the juxtaposition with the civil rights era since the last time a president did this, it was to push back against segregationists while this time it was about breaking up a conga line. Absent that context, it strips a radical encroachment on state sovereignty of its newsworthiness.

The algorithm also apparently wanted the article to tone down its characterization of Judge Breyer’s response:

Another line was flagged for suggesting Judge Charles Breyer had “pushed back” against the federal government in his ruling, an opinion which had called the president’s deployment of the National Guard the act of “a monarchist.” Rather than “pushed back,” the bias indicator suggested a milder word, like “disagreed.”

This new bot would have reported Watergate as a tenant association dispute.

In another example, BiasBot told Law360 that its coverage of a case should “state the facts of the lawsuit without suggesting its broader implications.” Given that the law is still ostensibly a function of precedent, reporting on caselaw is… all about broader implications.

It’s kind of the whole reason LexisNexis is in business, actually!

As a sometimes tech reporter, I have great relationships with the LexisNexis folks working to make the legal profession more efficient. But that’s because my contacts aren’t the people trying to micromanage news coverage to make sure every article earns the right-wing podcaster seal of approval as “fair.” It seems to me, the company might need to get control of its rogue unit.

There are, admittedly, opportunities to leverage generative AI in the journalist workflow. Detecting bias is not one of them for several reasons. The most straightforward and technical of which is that generative AI tools are designed to give the user pleasing answers come hell or high water. It’s how AI hallucinates cases to match the user’s research query. So if you build an AI to “detect bias” it guarantees that it will find some bias. Probably 4 or 5 bulleted examples no matter what. Does it really have a problem with “pushed back” or was that just something it grabbed to fill its answer quota?

But the more philosophical answer is that objective facts often have a lean. When 99 percent of climate scientists say climate change is real, do news outlets have to give equal time to Professor Daniel Plainview about the medicinal benefits of drinking crude oil? Because the algorithm can’t handle that nuance. Based on the examples in the NiemanLab piece, it’s just performing the barest level of sentiment analysis and flagging phrasing that carry even the slightest impact beyond the superficial. But that in and of itself is an act of bias. I used to tell deponents not to speculate because if they don’t know something — no matter how much they think they’re helping — they’re actually lying if they don’t admit that they don’t know.

The flip side is also true. A news report that says Charles Breyer had a tepid disagreement with the DOJ is, in fact, a lie. And it’s not any less of a lie because you asked the robot to say the lie for you.


HeadshotJoe Patrice is a senior editor at Above the Law and co-host of Thinking Like A Lawyer. Feel free to email any tips, questions, or comments. Follow him on Twitter or Bluesky if you’re interested in law, politics, and a healthy dose of college sports news. Joe also serves as a Managing Director at RPN Executive Search.

The post Law360 Using AI Bias Detector To Make Sure Stories Don’t Accidentally Tell The Truth appeared first on Above the Law.

GettyImages 2193585826

The biggest story in journalism right now is that CBS News agreed to give Donald Trump $16 million in a legally blessed bribe. The great sin of “The House That Edward R. Morrow Built” involved 60 Minutes airing a run-of-the-mill interview with Kamala Harris that made her look like a competent public servant with years of experience. Since Trump’s interviews, regardless of editing, sound like a dementia patient navigating a law school cold call, he decided CBS had committed consumer fraud because Harris spoke in complete sentences.

But apparently we weren’t done with today’s “dystopian assault on freedom of the press” news! And it came after an unlikely target: Law360. I certainly didn’t have “legal industry trade publication” on my censorship BINGO card. Then again, Biglaw lateral moves have suddenly become political stories so perhaps this marks inevitable cowardice creep reaching the legal press.

But the part of this story that elevates it from ominous development for civil liberties to comi-tragic is that Law360 is owned by LexisNexis and therefore the agent of Law360’s doom is… an AI algorithm! A new bias detecting ChatGPT wrapper slapped together by some LexisNexis product engineers probably taken away from actually useful work to build a degenerative AI to strip news articles of any semblance of value. 2025, man… Does. Not. Miss.

NiemanLab, Harvard’s digital journalism center, reports that Law360 has ordered its reporters run their stories through an AI bias detector designed for “applying a neutral voice to copy” and to be mandatory for “headline drafting, story tagging, and ‘article refinement and editing.’”

As one might imagine the journalists, represented by the Law360 union, object to this half-baked idea. A policy this ethically bankrupt could only arise from non-journalist executive input.

The announcement came a few weeks after an executive at Law360’s parent company accused the newsroom of liberal political bias in its coverage of the Trump administration. At an April town hall meeting, Teresa Harmon, vice president of legal news at LexisNexis, cited unspecified reader complaints as evidence of editorial bias.

Giving uncritical weight to squeaky wheel complaints, especially in an environment where a government official weaponized his followers to act on their every grievance up to and including STORMING THE FUCKING CAPITOL, is a dunderheaded management strategy only an MBA could come up with. But it’s almost certainly a cynical one. If we all start writing complaints that the headlines are neutered doublespeak, will Law360 be ordered to reverse course? I’m incredulous.

While the article notes that there’s not an established throughline from those remarks to the implementation of the policy, it speaks to a mindset that clearly got out of hand.

But let’s put aside the wisdom of the policy and focus on the fact that the bias detector is also terrible at its job. Because that’s just a little bit more fun. Only at a tech company could someone think that generative AI tools being developed for dedicated legal work tasks could be bolted onto the editorial process of a news publication.

Generative AI is a powerful tool in the same way a screwdriver is a powerful tool. But you wouldn’t use a screwdriver to do your taxes. Yet that’s the thinking involved in bringing AI into an editorial process. To borrow from the TV series Veep, it’s like using a croissant as a dildo: “It doesn’t do the job, and it makes a fucking MESS!”

She also criticized the headline of a March 28 story — “DOGE officials arrive at SEC with unclear agenda” — as an example. In the same town hall, Harmon suggested that the still experimental bias indicator might be an effective solution to this problem, according to two employees in attendance.

But… DOGE officials did arrive at the SEC with an unclear agenda. The White House couldn’t be clear about who was running DOGE let alone its agenda. This is just a factual statement that, if anything, is biased in favor of DOGE since its suspected agenda to steal data and hamper regulation was about as disguised as three raccoons in a trench coat.

The report notes another story about the Trump decision to mobilize the California National Guard:

Several sentences in the story were flagged as biased, including this one: “It’s the first time in 60 years that a president has mobilized a state’s National Guard without receiving a request to do so from the state’s governor.” According to the bias indicator, this sentence is “framing the action as unprecedented in a way that might subtly critique the administration.” It was best to give more context to “balance the tone.”

It was the first time in 60 years though! That is the relevant context. As is the juxtaposition with the civil rights era since the last time a president did this, it was to push back against segregationists while this time it was about breaking up a conga line. Absent that context, it strips a radical encroachment on state sovereignty of its newsworthiness.

The algorithm also apparently wanted the article to tone down its characterization of Judge Breyer’s response:

Another line was flagged for suggesting Judge Charles Breyer had “pushed back” against the federal government in his ruling, an opinion which had called the president’s deployment of the National Guard the act of “a monarchist.” Rather than “pushed back,” the bias indicator suggested a milder word, like “disagreed.”

This new bot would have reported Watergate as a tenant association dispute.

In another example, BiasBot told Law360 that its coverage of a case should “state the facts of the lawsuit without suggesting its broader implications.” Given that the law is still ostensibly a function of precedent, reporting on caselaw is… all about broader implications.

It’s kind of the whole reason LexisNexis is in business, actually!

As a sometimes tech reporter, I have great relationships with the LexisNexis folks working to make the legal profession more efficient. But that’s because my contacts aren’t the people trying to micromanage news coverage to make sure every article earns the right-wing podcaster seal of approval as “fair.” It seems to me, the company might need to get control of its rogue unit.

There are, admittedly, opportunities to leverage generative AI in the journalist workflow. Detecting bias is not one of them for several reasons. The most straightforward and technical of which is that generative AI tools are designed to give the user pleasing answers come hell or high water. It’s how AI hallucinates cases to match the user’s research query. So if you build an AI to “detect bias” it guarantees that it will find some bias. Probably 4 or 5 bulleted examples no matter what. Does it really have a problem with “pushed back” or was that just something it grabbed to fill its answer quota?

But the more philosophical answer is that objective facts often have a lean. When 99 percent of climate scientists say climate change is real, do news outlets have to give equal time to Professor Daniel Plainview about the medicinal benefits of drinking crude oil? Because the algorithm can’t handle that nuance. Based on the examples in the NiemanLab piece, it’s just performing the barest level of sentiment analysis and flagging phrasing that carry even the slightest impact beyond the superficial. But that in and of itself is an act of bias. I used to tell deponents not to speculate because if they don’t know something — no matter how much they think they’re helping — they’re actually lying if they don’t admit that they don’t know.

The flip side is also true. A news report that says Charles Breyer had a tepid disagreement with the DOJ is, in fact, a lie. And it’s not any less of a lie because you asked the robot to say the lie for you.


HeadshotJoe Patrice is a senior editor at Above the Law and co-host of Thinking Like A Lawyer. Feel free to email any tips, questions, or comments. Follow him on Twitter or Bluesky if you’re interested in law, politics, and a healthy dose of college sports news. Joe also serves as a Managing Director at RPN Executive Search.