VIEWS1.1KBOOKMARKS2REPLIES2
Ben Golub@ben_golub
Full @pangramlabs analysis - the gold standard in AI detection
https://www.pangram.com/history/d93f9a80-8922-4f85-b3c9-51159a6079e2?ucc=z9Wh3mgWVbt
28dViews 1.1KLikes 2Bookmarks 2
Pangram Labs flags 48 percent AI text in Oklahoma ruling
Pangram Labs analysis determined that 48 percent of the text in a magistrate judge opinion was AI-generated. The May 13 2026 filing in case 6:25-cv-00227-JAR granted a preliminary injunction in Muscogee (Creek) Nation v. City of Henryetta in the Eastern District of Oklahoma. The document cites McGirt v. Oklahoma and Hooper v. City of Tulsa on tribal sovereignty and was shared publicly with the detection report.
2010223423.8K
Original post unavailable.
Sentiment
Users dismissed claims that an Oklahoma federal magistrate's opinion was half-written by AI, arguing that detectors like Pangram are unreliable for legal texts and prone to errors from formatting or markdown.
Pos
0.0%
Neg
100.0%
4 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
LIKES5
Itai Sher@itaisher
@ben_golub @fedjudges I think the assessment of this depends on what 48% AI composed means.
Is it that many ideas are due to AI or that the judge wrote an initial draft and had AI smooth over the wording?
Really the question is legitimate use and I am not sure these detectors are geared for that.
28dViews 207Likes 5
Ben Golub@ben_golub
@pangramlabs Opinion pdf
https://storage.courtlistener.com/recap/gov.uscourts.oked.38772/gov.uscourts.oked.38772.85.0.pdf
28dViews 972Likes 2Bookmarks 1
Delta Vee@deltaVee42
@ben_golub @fedjudges CC @RobertFreundLaw.
28dViews 109Likes 1Bookmarks 1
Kara 🦇 🔊@0xkarasy
@ben_golub @pangramlabs @MathpixApp You can clearly see that Pangram claims 12x (whatever that means) just because you used markdown.
I didn’t read the submission yet, all I’m saying is whether it is Turnitin AI or Pangram, these tools are unreliable to tag as evidence for AI slop.
28dViews 70Likes 2
Jaden Halstead@JadenHalstead
@ben_golub @fedjudges I’m curious as to how people still think AI checkers have any amount of accuracy
28dViews 90Likes 1
Kara 🦇 🔊@0xkarasy
@ben_golub @pangramlabs You made a few mistakes, too.
Why did you copy as markdown?
28dViews 16Likes 1
Auyon Siddiq@auyonomous
@ben_golub @fedjudges Was confused for a moment why Rotten Tomatoes is reviewing judge opinions
28dViews 224Likes 2
Ben Golub@ben_golub
@0xkarasy @pangramlabs I converted pdf to markdown using the best tool for that, @MathpixApp, so I could get a faithful conversion of the text without headers or footers interfering.
If you know AI slop, you'll recognize this as very much AI slop.
28dViews 21
Kara 🦇 🔊@0xkarasy
@ben_golub @fedjudges Legal writing is notoriously difficult for AI Slop testers to evaluate for AI slop. Pangram is really bad at checking my legal writing. Sometimes it says 100% human and next check of the text, 40-50% AI. These aren’t reliable results.
28dViews 128Likes 2
Karl Rohe@karlrohe
@ben_golub @fedjudges Would probably be wise to compare this result against the judge’s previous opinions. To me, this reads like ai. But possibly just their style and we likely have the data to tell.
28dViews 199Likes 1
Valeriy M., PhD, MBA, CQF@predict_addict
@ben_golub @fedjudges Fake news, there are no technologies that can reliably tell whether a text has been AI generated.
28dViews 221
Ben Golub@ben_golub
@JadenHalstead @fedjudges You should really play around with Pangram (it's free to try) and tell me your experience.
28dViews 84
Pangram Labs@pangramlabs
@0xkarasy @ben_golub @MathpixApp Hi there! The 12x here refers to 12x more likely to appear in AI writing vs. human. It's part of our Supporting Evidence feature, and isn't used in detection.
28dViews 10Likes 1
Pangram Labs@pangramlabs
@ben_golub Extracted 537 words from the link.
We believe that this document is fully human-written
https://www.pangram.com/history/5bf4883e-cbde-4354-9203-78f75dd9c0f3
28dViews 31