Pangram's AI scores are fueling public shaming campaigns
AI-text-detection company Pangram hired journalist Rod Breslau to publicly shame social media users suspected of using AI, effectively acting as what the piece calls an "attack dog." According to WIRED, Breslau offered his services to Pangram founder Max Spero earlier this year because he was tired of seeing AI-generated content online; he then advised the company on social media strategy while actively hunting down suspected AI users. "I wanted to take a harsher approach because I thought that people were getting off too easy," he told WIRED. In an earlier interview with GamesBeat, Breslau argued that anyone who builds a personal brand as a thought leader on Twitter or LinkedIn, including executives and tech CEOs, "should have to write your own posts," and that anyone who doesn't deserves to be mocked for it.
Pangram later cut ties with Breslau as the company changed direction; per WIRED, it now expects AI use to become increasingly accepted and wants to detect even light AI editing, positioning itself as the authority for originality checks across publishing, education and beyond. But dropping Breslau hasn't stopped the shaming: Spero has kept publicly scoring people and calling out alleged AI use, framing it as holding people accountable when they try to hide AI involvement.
The article's central argument is that Pangram, and the detection scores it sells, only measure whether AI was likely involved in a piece of writing, never how. Even a perfectly accurate score can't tell whether someone generated an entire text from a prompt, used AI as a thinking partner, asked it for help with wording, or simply ran an already-written draft through it to smooth the language or translate it. Yet the public shaming built on those scores assumes any AI involvement means the writer didn't think or didn't care, treating a text built from hours of original research the same as one prompted into existence in ten seconds.
The author illustrates this with their own case: the English version of the article is a translation of a German original, drafted with AI help and then edited by hand and by AI throughout. Pangram scored it "28 percent AI" and flagged only the final paragraphs as AI-written, even though every section of the piece was produced and edited the same way, with roughly the same amount of AI involvement and human intention. The author adds, from their own unspecified testing, that Pangram's scores are often not fully accurate. The stakes go beyond social media call-outs: the piece says academic institutions are already rejecting papers based on percentage AI-detection scores, penalizing researchers who write in a second language or who are simply better at science than at prose and use AI to state their results more clearly.
The article closes on a historical parallel and a business-model critique. Novelist Alexandre Dumas had assistants such as Auguste Maquet write rough drafts of his novels, including The Three Musketeers, which Dumas then refined himself; an AI detector applied to Dumas, the piece argues, would have flagged "Maquet usage" and missed everything that made the books what they are. It also argues that Pangram's business model depends on the public equating AI use with laziness or dishonesty, since less stigma around AI would leave less reason to pay for detection in the first place.
Key facts
- Pangram hired journalist Rod Breslau, described as an "attack dog," to publicly shame social media users suspected of using AI, then cut ties with him as the company changed direction.
- Even after dropping Breslau, Pangram CEO Max Spero has kept publicly scoring people and calling out alleged AI use based on the company's own detection results.
- Pangram's detection score measures only whether AI was likely involved, not how; it cannot tell an entirely AI-generated text apart from one where AI just helped with wording, editing or translation.
- Pangram scored the English version of this very article, a translation of a German original drafted and edited with a mix of AI and human work throughout, at "28 percent AI" and flagged only its final paragraphs.
- The piece warns that academic institutions already reject papers based on percentage AI-detection scores, penalizing researchers who write in a second language or who use AI to sharpen their prose.
Why it matters
The piece's core argument is that AI-detection tools like Pangram answer only one question, whether AI touched a piece of text, while the public shaming built on their scores answers a completely different one: whether the person behind it actually thought or worked for themselves. Conflating the two lets a score built from hours of original research get treated the same as one prompted into existence in seconds. The article also argues that Pangram's business itself depends on that conflation: if the stigma around AI use fades, so does the reason to pay for detection, which gives the company an incentive to keep the two questions blurred even as its own stated strategy shifts toward detecting light AI editing everywhere from publishing to education.
Who it affects
Directly, it affects social media users whom Breslau and Spero have singled out and called out publicly using Pangram's scores, plus anyone who might be targeted the same way going forward. More broadly, it affects researchers and academics: the piece says institutions are already rejecting papers based on percentage AI-detection scores, which lands hardest on people writing in a second language, or on scientists who are stronger at their research than at English prose and lean on AI to state results clearly. It also affects the author of the piece directly, whose own translated, partly AI-edited article was scored and partially flagged by the tool it critiques.
How to use it
The piece's practical lesson is about how to read an AI-detection score rather than how to use Pangram itself: a percentage score says only that AI was likely involved somewhere, not that the writer failed to think, failed to do original work, or tried to deceive anyone. Before treating a score as evidence of laziness or dishonesty, whether on social media or in an academic review, the piece argues you would need to know how AI was used: to draft from scratch, to think alongside the writer, to fix wording, or just to translate or smooth an already-finished text, distinctions no current score reports.
How solid is it
This is an analysis and opinion piece published on The Decoder, written in the first person by an author the text never names. Its reported core, that Pangram hired Breslau and later cut ties with him while Spero kept up the public scoring, is credited to WIRED and includes direct, on-the-record quotes from Breslau himself, to WIRED and in an earlier GamesBeat interview. The claim that Pangram's scores are "often" inaccurate rests on the author's own unspecified testing, and the claim that academic institutions reject papers over percentage scores is stated without naming an institution, a paper count, or a figure. The Dumas-and-Maquet comparison is presented as the author's own analogy, not a claim from Pangram, Breslau or Spero.
Risks and caveats
The piece leaves several specifics unstated: no date is given for when Breslau approached Pangram or when WIRED's report ran, beyond "earlier this year"; no detail is given on why Pangram "changed direction" after cutting ties with Breslau, or on any compensation or formal title behind his work for the company; and the claim that Pangram's own scores are often inaccurate, along with the claim about academic institutions rejecting AI-flagged papers, are both stated as assertions rather than backed with named examples, counts or figures. The author's identity is also never given in the text, despite the piece being written in the first person about their own testing and their own translated article.
“I wanted to take a harsher approach because I thought that people were getting off too easy”
— Rod Breslau, to WIRED