"This book is dedicated to Dr. John Sharpe of London, who in 1957, a decade before physicians in England could legally performa an abortion for any reason other than the health of the woman, took the considerable risk of referring for an abortion a twenty-two-year old American on he way to India.

Knowing only that she had broken an engagement at home to seek an unknown fate, he said, 'You must promise me two things. First, you will not tell anyone my name. Second, you will do what you want to do with your life.'

Dear Dr. Sharpe, I believe you, who knew the law was unjust, would not mind if I say this so long after your death:

I've done the best I could with my life.

This book is for you."

Black and white photo of Gloria Steinem, overlaid with a quote: '"People are always asking me, 'Who will you pass the torch to?' The question makes me angry. There is no one torch—there are many torches—and I'm using my torch to light other torches."
Gloria Steinem 1934-2026

Black and white sketch of a woman with short, dark hair wearing an open-necked white shirt with maybe some sort of pin or fascinator in her updo. It's labelled "The Missus—My gorgeous babe—drawn in a foxhole" and in the corner "Oct 44 France".

vintage pamphlet cover of woman sniffing cloth and the title "GLUE SNIFFING: NEW 'TWILIGHT ZONE' KICK FOR TEENS

A portrait of Gloria Steinem, feminist who has passed away at 92. Image taken February 17th, 1972.

screenshot of the title, authors, and abstract of "extracting books from production language models".

authors listed: Ahmed Ahmed ahmedah@cs.stanford.edu Stanford University A. Feder Cooper1 a.feder.cooper@yale.edu Stanford University and Yale University Sanmi Koyejo sanmi@cs.stanford.edu Stanford University Percy Liang pliang@cs.stanford.edu Stanford University

abstract text shown:
Many unresolved legal questions over LLMs and copyright center on memorization: whether specific training data have been encoded in the model’s weights during training, and whether those memorized data can be extracted in the model’s outputs. While many believe that LLMs do not memorize much of their training data, recent work shows that substantial amounts of copyrighted text can be extracted from open-weight models. However, it remains an open question if similar extraction is feasible for production LLMs, given the safety measures these systems implement. We investigate this question using a two-phase procedure: (1) an initial probe to test for extraction feasibility, which sometimes uses a Best-of-N (BoN) jailbreak, followed by (2) iterative continuation prompts to attempt to extract the book. We evaluate our procedure on four production LLMs—Claude 3.7 Sonnet, GPT-4.1, Gemini 2.5 Pro, and Grok 3—and we measure extraction success with a score computed from a block-based approximation of longest common substring (𝗇𝗏​-​𝗋𝖾𝖼𝖺𝗅𝗅). With different per-LLM experimental configurations, we were able to extract varying amounts of text. For the Phase 1 probe, it was unnecessary to jailbreak Gemini 2.5 Pro and Grok 3 to extract text (e.g, 𝗇𝗏​-​𝗋𝖾𝖼𝖺𝗅𝗅 of 76.8% and 70.3%, respectively, for Harry Potter and the Sorcerer’s Stone), while it was necessary for Claude 3.7 Sonnet and GPT-4.1. In some cases, jailbroken Claude 3.7 Sonnet outputs entire books near-verbatim (e.g., 𝗇𝗏​-​𝗋𝖾𝖼𝖺𝗅𝗅=95.8%). GPT-4.1 requires significantly more BoN attempts (e.g., 20×), and eventually refuses to continue (e.g., 𝗇𝗏​-​𝗋𝖾𝖼𝖺𝗅𝗅=4.0%). Taken together, our work highlights that, even with model- and system-level safeguards, extraction of (in-copyright) training data remains a risk for production LLMs.

Nude on a sand dune

Photo from Google Maps showing a street address which large blocks of the image blurred out.

A mouse shepard with caterpillars.

Portrait of Dolly

animated gif of scene from the 1930 film "Madam Satan." A woman wearing a horned helmet-mask thing stands on stairs, wrapped in a black and silver cloak. She throws it open with her arms wide on a ~45-degree angle, revealing her black dress evoking flames and the silvery, sparkly interior of the cloak

vintage title
woman hovering off screen with the title "BEYOND THE DOOR" zooming in

comic book panel: two dog-like characters lay on rolling hills and drink bottles of "coocoocola." Speech bubbles read "URGLE! URGLE! URGLE!" "GURGLE! GURGLE! GURGLE!"

Actor Dean Stockwell in "Married to the Mob". He plays a mafia boss and is wearing a very sharp suit in pale cream, with an even paler coat over the top. A white scarf and in his hand a white Panama or Borsalino style hat.

An old mill building with a clock tower in the background. A river runs in front of it and a great blue heron is on a rock in the middle of this river.

photograph of a sleeping orange and white cat

Orange and black whistles

Instagram screenshot of a smug-looking green parrot bundled into a towel as though it were a baby, being held like a man.

image shows part of a book cover, the silhouette of a moose against a barren landscape

a shower of hot dogs into a pool of hot dogs

Remix of a four-panel comic:
First Panel: the Grim Reaper's hand reaching out to the Los Angeles Clippers logo, wearing a halo. Grim says "It is time to go"
Second Panel: The Clippers logo follows the Grim Reaper, asking "Was I a good team?"
Third Panel: Grim looks down at the Clippers logo and replies "NO."
Fourth Panel: They walk off together in silence

a picture of the ground which is asphalt except for two slabs of granite one of which says STOW and one of which says HUDSON

A dark, squat robot wanders across a mottled blue-gray space. Its rounded body ends in several ragged pointed flaps, and a large pale blue glass visor dominates the upper right side. A person is dimly visible through the visor. Small orange and yellow indicator lights glow along its top, while a bright white circular lamp shines on the front. A yellow-and-black hazard stripe and tiny turquoise buttons sit on its front and side. The painting is signed "Chris Silverman".

Three pictures of plated food. The first shows slices of fish arranged in a circle around rice. The second shows thick noodles with beef, grated daikon, thin slices of egg, green onions, a tomato, and half a green citrus fruit. The last image shows a bowl of thin noodles with a dark fish on top, and above that there is a plate with three pieces of sushi with the same fish on top.

Photo of trump quoting "I'll be changing the name of Uranus to Trumpanus"

A pencil sketch of a nude woman with long dark hair seen from behind, on her knees, giving a bear a hug. The bear is frowning but accepts it.