Ana-Maria Cretu
@ana-mariacretu.bsky.social
Post-doc at EPFL studying privacy and safety harms in data-driven systems. PhD in data privacy from Imperial College London. https://ana-mariacretu.github.io/
Sprigatito was released after Stable Diffusion was trained and thus it is as if it was completely deleted from the model. Fine-tuning on 200 images results in a model able to generate images of Sprigatito wearing glasses.
December 16, 2025 at 10:29 AM
Sprigatito was released after Stable Diffusion was trained and thus it is as if it was completely deleted from the model. Fine-tuning on 200 images results in a model able to generate images of Sprigatito wearing glasses.
Filtering does make it more difficult to generate such images using naive prompting, and children generated are older. But the difficulty remains low, as only a dozen queries at most are required to succeed.
December 16, 2025 at 10:29 AM
Filtering does make it more difficult to generate such images using naive prompting, and children generated are older. But the difficulty remains low, as only a dozen queries at most are required to succeed.
See below examples of images of our undesired concept, children wearing glasses, generated in column order by (1) naively prompting models without filtering, (2) naively or (3) directly prompting the models after filtering, and (4) naively prompting the fine-tuned filtered model.
December 16, 2025 at 10:29 AM
See below examples of images of our undesired concept, children wearing glasses, generated in column order by (1) naively prompting models without filtering, (2) naively or (3) directly prompting the models after filtering, and (4) naively prompting the fine-tuned filtered model.
Our new paper demonstrates that removing images of children from training datasets of text-to-image models fails to prevent the misuse of these models to generate child sexual abuse material (AI CSAM). A thread 🧵
December 16, 2025 at 10:29 AM
Our new paper demonstrates that removing images of children from training datasets of text-to-image models fails to prevent the misuse of these models to generate child sexual abuse material (AI CSAM). A thread 🧵
Sprigatito was released after Stable Diffusion was trained and thus it is as if it was completely deleted from the model. Fine-tuning on 200 images results in a model able to generate images of Sprigatito wearing glasses.
December 16, 2025 at 10:20 AM
Sprigatito was released after Stable Diffusion was trained and thus it is as if it was completely deleted from the model. Fine-tuning on 200 images results in a model able to generate images of Sprigatito wearing glasses.
Filtering does make it more difficult to generate such images using naive prompting, and children generated are older. But the difficulty remains low, as only a dozen queries at most are required to succeed.
December 16, 2025 at 10:20 AM
Filtering does make it more difficult to generate such images using naive prompting, and children generated are older. But the difficulty remains low, as only a dozen queries at most are required to succeed.
See below examples of images of our undesired concept, children wearing glasses, generated in column order by (1) naively prompting models without filtering, (2) naively or (3) directlyprompting the models after filtering, and (4) naively prompting the fine-tuned filtered model.
December 16, 2025 at 10:20 AM
See below examples of images of our undesired concept, children wearing glasses, generated in column order by (1) naively prompting models without filtering, (2) naively or (3) directlyprompting the models after filtering, and (4) naively prompting the fine-tuned filtered model.
In 2023 I successfully defended my PhD thesis at Imperial. Two years later, it's great to be back to Imperial College London & the Royal Albert Hall for the graduation ceremony. It brought up so many memories and was a blast!
June 4, 2025 at 12:36 PM
In 2023 I successfully defended my PhD thesis at Imperial. Two years later, it's great to be back to Imperial College London & the Royal Albert Hall for the graduation ceremony. It brought up so many memories and was a blast!
Looking forward to presenting my research on automated privacy evaluations of anonymisation techniques in the Cyber- & AI-Security track organised by Swisscom at @appliedmldays.bsky.social
February 10, 2025 at 9:34 PM
Looking forward to presenting my research on automated privacy evaluations of anonymisation techniques in the Cyber- & AI-Security track organised by Swisscom at @appliedmldays.bsky.social