An article about me on Grokipedia is full of disinformation in which the LLM strained to fulfill its programming mandate to present a conservative viewpoint.
[See this Feb. 1, 2026 post for an update.]
“Grokipedia,” Elon Musk’s AI-written alternative to Wikipedia that is touted as de-wokified information, recently came to my attention, so I (naturally) looked up my entry. The basic details of my biography were accurate. The content appeared to be lifted from what Grok found on Wikipedia, my bio page on this website, my reporter page on The New York Times website and similar sources – same essential facts, with some light and superficial rewriting, like a high school student’s opsec for plagiarism. And it was longer, pulling in many minor additional details Grok found on other websites that were also accurate – what level of honors was my undergraduate degree, if anyone cares, for instance. I found it refreshing that Grok omitted my teenaged connection to “The Dark Side of the Rainbow,” which people like to bring up but has little to do with my actual professional life and role.
But where the results had political overtones — like a “Controversies and Criticism” section — the system invented articles that do not exist and twisted real-world episodes in (like two instances in which I criticized right-wing outlets for publishing false factual claims, both of which resulted in concessions of error on their part) into fictionalized narratives about bias. My assessment is that the Grok algorithm is so overtorqued to come up with critical things to say from a politically conservative point of view about any topic it covers, that where there is insufficient substance to work with, it strains to invent something.
Here are the things it came up with, and why they are disinformation:
National Review / Kevin Clinesmith 2020
Fake news! There was no such July 30 article by me or August 3 article by McCarthy. And while there was a critique related to Clinesmith roughly in that time period, it went the other way.
In the real world, on August 19, Clinesmith pleaded guilty to a false statement charge for having doctored a CIA email during the fourth and final Carter Page FISA application in mid 2017; Clinesmith’s crazy act kept the fact that all four applications omitted a mitigating fact — Page used to talk to the CIA about his business interactions with Russians, before the agency cut him off following a 2013 GRU recruitment effort — from coming to internal light inside the FBI. Andy McCarthy wrote a multipart series for the National Review on Aug. 24, 25 and 26 about the Clinesmith case. The first two parts of his take mangled certain facts. Knowing the details because I had covered the case – indeed, with my colleague Adam Goldman, I co-wrote the first story breaking the news of Clinesmith’s actions and naming him – I wrote a tweet thread on August 24 and 25 that dissected the factual distortions in the National Review.
Footnote 43 in the Grokipedia article, the supposed source for the notion that “publications like National Review have challenged [my] interpretations of events such as the Kevin Clinesmith case,” actually goes to McCarthy’s response to my critique. In it, he said some nice things about me. He contended that some of the problems I pointed out in his pieces were “overwrought.” But he conceded error on the main point and wrote that “[Savage] was quite right to call me on it” and said he was correcting it.
I never wrote that Clinesmith’s act, or the larger omission of Page’s prior relationship with the CIA in the warrant applications, did not materially affect the FISA warrant’s validity. And none of this involved McCarthy criticizing my coverage of the Clinesmith case. He never accused me of downplaying the FBI’s botching of the FISA applications as just an isolated error by one bad apple (indeed, my coverage focused on systemic problems in both those applications and the FBI’s approach to FISA even more broadly – see this and this and this). McCarthy also never accused me of selectively quoting from the plea agreement. Grok just made all of that up, along with inventing articles by both of us that do not exist, to create a funhouse-mirror version of this episode in which the National Review was criticizing me when it was the other way around.
Washington Examiner / Durham investigation 2022
Fake news redux! This is all hallucinated. None of the supposed articles exist; this never happened. Nobody accused me of prematurely dismissing Durham’s findings based on incomplete information. The Times does not even have a public editor.
But I think I can dimly perceive the origin through the fog of disinformation, and once again the direction goes the other way. While the actual Durham news in May 2022 was the trial and acquittal of Michael Sussmann, a few months earlier I was involved in a Durham-related fracas with a different writer at the Washington Examiner, and the LLM must have drawn on that to invent this fictional tale.
The Washington Post wrote about the real-world episode here. But to summarize: In February 2022, a poorly written pretrial court filing by Durham set off an uproar among Trump and right-wing media outlets, who claimed it said Durham had found evidence of “spying” on Trump White House internet data by a tech executive who, through Sussmann, had connections to Democrats. This was all wrong; among other important things, the White House internet data in question (indicating that a Russian-made smartphone rarely seen in the United States had been connecting to the White House’s wi-fi network) actually came from when Obama was president.
When Trump and his allies started shaming the mainstream media for engaging in a cover-up because actual journalistic outlets were not repeating their fevered fantasy (Trump: “The press refuses to even mention the major crime that took place. This in itself is a scandal, the fact that a story so big, so powerful and so important for the future of our nation is getting zero coverage from LameStream, is being talked about all over the world.”), I wrote a news analysis article explaining how the entire narrative was simply off track. I cited as one example a Washington Examiner article with the inaccurate headline “Durham says Democrat-allied tech executive spied on Trump’s White House office.”
The next day, the Examiner‘s commentary editor, Conn Carroll, wrote a clumsy attempted hit piece that doubled down on the Trump White House data claim while also libelously attacking me with a nasty headline containing my name and supported by two blatant lies. For example, Carroll wrote that I had concealed that Sussmann had ties to Democrats, “calling him merely a ‘cybersecurity lawyer,'” when I had literally described Sussmann as a “cybersecurity lawyer with links to the Democratic Party.”
After Durham acknowledged in a subsequent court filing that he had indeed been talking about Obama-era White House data, I used Twitter to ask the Examiner to retract its lies about me. When that was ignored, a defamation lawyer hired by the New York Times sent two letters. The Examiner shadow-edited the article to tone down how out on a limb Carroll had gone about the Trump-era claim, and it did correct one of Carroll’s lies, but not the other, while appending an editor’s note of regret:
That’s it. I am not aware of anything else – and find nothing on Google — that supports a claim that the Examiner demanded I retract coverage about Durham. The material Grokipedia is presenting is fiction.
In sum, straining to point to something substantive that would support a conservative view of bias, Grok has invented articles that do not exist and warped an episode in which I demanded corrections/retractions from the Washington Examiner into one where it demanded them from me.
Power Wars was too short = political bias?
This is a big time garble, too.
The paragraph about the book cites a 2017 blog post by Charlie Dunlap, a Duke law professor and former senior Air Force JAG, that lamented that Power Wars, my 800-page book about national security legal policymaking inside the Obama administration, did not delve much into what career military lawyers had been up to during the Obama years.
This was true. I had focused on the actions and deliberations of politically appointed lawyers because I wanted to write a book exploring and dissecting the accusation that Obama had, contrary to the expectations raised by his 2008 campaign rhetoric, ended up “acting like Bush” on many post-9/11 issues.
Dunlap clearly enjoyed the book for what it was and just wished I had made the already long book even longer by adding some similar investigative/explanatory journalism about contemporaneous JAG matters that he, personally, was also very interested in. He had a tart line in the middle of his blog post about my choice not to write much about JAGs: “Power Wars is one where the author has a conclusion that discussing the machinations of civilian, policy-making lawyers better supports his contentions. In other words, Power Wars is essentially a polemic, not an impartial and comprehensive legal history of the period.”
The paragraph loosely derived from Dunlap’s post and citing to it is the top example Grokipedia has for concocting a section about “perceived bias in coverage.” So it is interesting that Dunlap himself addresses whether my omission of JAG content from this era amounts to bias. Here is what he said: “Does this mean Savage has some anti-military or anti-JAG bias? I don’t think so. To my knowledge, he has not demonstrated any hostility or even dismissiveness about the role or function of JAGs. Quite the contrary, the efforts of uniformed lawyers to resist the excesses of the civilian government lawyers during the Bush years were prominently featured in his 2007 book, Takeover: The Return of the Imperial Presidency.”
Indeed, I had written a fair bit about JAGs in my Bush-Cheney book because in the early years after 9/11, they kept getting into big fights with politically appointed lawyers over detainee treatment and interrogation policies that the career uniformed lawyers thought violated the Geneva Conventions. There just was not equivalent JAGs vs. politicals fighting in the Obama years to write about. But there were plenty of other Obama-era national-security legal policy and executive power controversies to investigate and explain, which is why the book was (too) long.
The important thing here is that Dunlap never said or implied that the lack of JAG material in my Obama book relative to my Bush-Cheney book demonstrated “bias favoring the Obama administration’s national security policies” — that just wasn’t the point he was making at all. Grokipedia contorted his hobbyhorse critique into a politicized non sequitur.
A Sourceless Claim of Asymmetric Scrutiny
The paragraph about my New York Times reporting supposedly showing asymmetrical scrutiny, focusing more on Republican administration abuses than Democratic ones, has no source at all. It also seems to misportray my old Pulitzer Prize-winning series on signing statements and executive power in The Boston Globe – not The Times, as Grok implies – as if it were not about legal rationales and internal deliberations, which it very much was. (It sounds like Grok is maybe confusing my series with two that won prizes the previous year for exposing warrantless wiretapping and detainee controversies.) I’m not surprised there’s no source for this one, because as the Obama and Biden legal teams will tell you, I scrutinized their controversial policies about drone strikes, their expediently shifting views of unilateral presidential war powers, their leak prosecutions and all kinds of other executive power and national security legal policy issues more than any other reporter at any outlet.
Noticing the gap between the hype and the results of Durham probe is bias?
It’s true I pointed out the gap between the political expectations about the Durham probe and its results. So did Trump allies Tom Fitton of Judicial Watch (“Durham let down the American people with few and failed prosecutions”) and former GOP Rep. Matt Gaetz (“When you are part of the cover-up, Mr. Durham, then it makes our job harder. … I think you let the country down.”) Right-wing media had stoked expectations for years that Durham was about to prove that there had been a deep state conspiracy and put high-level officials in prison. Instead, he developed two false-statements cases against non-government officials involved in outside efforts to investigate Trump and Russia, both of which ended in acquittals. His 2023 final report rehashed the substance of problems already uncovered in 2019 by an inspector general (including the Clinesmith issue), recycled some insinuations he had put into the indictments of his own failed cases even though he never found evidence sufficient to prove them, and accused the FBI of “confirmation bias” – not even political bias.
Grokipedia’s two cited purported sources for a paragraph declaring that that my pointing out this gap drew accusations of harsher treatment are the Durham report itself and a 2022 CNN column, neither of which make that accusation. The Durham report does not mention me; the CNN piece was about the false spying-on-Trump-White-House hullabaloo and favorably cited my dissection of how Durham’s court filing was being misinterpreted. But Grok’s algorithm required it to say something, so say something it did.
Conclusion
My takeaway is that Grokipedia might be useful for non-politicized topics — its material on the basics of my biography is more detailed than Wikipedia and is accurate. But it is factually unreliable when its programming mandate to include politically conservative observations kicks in: Grok demonstrably invents falsehoods as it strains to comply.