iiinova — where your digital self keeps evolving

Ana-Maria Cretu @ana-mariacretu.bsky.social · 24/08/2026

📣 We have extended the deadline to submit papers to our workshop until Aug 27th AOE! We are also looking for reviewers: if you are interested, please apply before Aug 26th AOE.

010

Ana-Maria Cretu @ana-mariacretu.bsky.social · 11/08/2026

🎉 We are excited to announce the 2nd Workshop on Foundations of Language Model Security, co-located with NeurIPS 26 in Paris, France!

110

Reposted by Ana-Maria Cretu

CISPA Helmholtz Center for Information Security @cispa.de · 15/07/2026

Can removing images of children from AI training datasets prevent text-to-image models from generating CSAM? New research by CISPA-Faculty Dr. @ana-mariacretu.bsky.social shows that dataset filtering provides only limited protection & can be circumvented. cispa.de/en/cretu-dataset-filtering

cispa.de

Dataset Filtering Provides Only Limited Protection Against CSAM Generation in Text-to-Image Models

CISPA-Faculty Dr. Ana-Maria Cretu investigated whether removing images of children from training datasets is an effective method.

031

Ana-Maria Cretu @ana-mariacretu.bsky.social · 22/05/2026

📢 Are you a researcher working on understanding and mitigating the harms of AI-generated non-consensual images of adults and children? Then our CARA '26 workshop (co-located with @soupsconference.bsky.social) taking place on Aug 24th in Hannover, Germany is for you!

120

Ana-Maria Cretu @ana-mariacretu.bsky.social · 22/05/2026

It was an honor to present our paper "Evaluating Concept Filtering Defenses against Child Sexual Abuse Material Generation by Text-to-Image Models" at the IEEE Symposium on Security and Privacy in San Francisco, California 😊 📜 @sp.ieee-security.org

160

Reposted by Ana-Maria Cretu

COSIC @cosic.bsky.social · 19/03/2026

Research by Maxime Deryck (UGent), Diane Leblanc-Albarel and Bart Preneel (COSIC, KU Leuven) is featured in CyberInsidercom: "Microsoft #PhotoDNA is vulnerable to false positives and data leakage". cyberinsider.com/microsoft-ph...

cyberinsider.com

Microsoft PhotoDNA is vulnerable to false positives and data leakage

Flaws in Microsoft’s PhotoDNA show it can be bypassed, manipulated to produce false matches, and partially reversed to reveal image content.

021

Ana-Maria Cretu @ana-mariacretu.bsky.social · 24/02/2026

I'm excited to announce that my research group at CISPA @cispa.de is recruiting PhD students to work on privacy, security, and user safety in general-purpose data systems, with a focus on generative AI systems, synthetic data, and anonymization. Are you interested in:

130

Reposted by Ana-Maria Cretu

CISPA Helmholtz Center for Information Security @cispa.de · 28/01/2026

Call for Papers: MetaCRiSP Workshop (Security & Privacy). A venue for metascience and critical reflection on research and peer-review practices in security & privacy. Workshop: May 21, 2026 · San Francisco. (47th IEEE S&P) Deadline: Feb 12, 2026. More here: metacrisp.org @mu00d8.bsky.social

112

Reposted by Ana-Maria Cretu

strufe @strufe.bsky.social · 27/01/2026

💥The year starts with great news:💥 🍾Àlex' paper accepted at USENIX Sec🎉 He demonstrates how suppression (and: sampling as one special case) may amplify privacy but yields disproportionate utility loss. It's been hard work: 🚀per aspera ad astra🌟 Congrats, we're proud! arxiv.org/abs/2601.05180

Screenshot of the title page of Alex' paper - "The Adverse Effects of Omitting Records in Differential Privacy: How Sampling and Suppression Degrade the Privacy–Utility Tradeoff"

063

Ana-Maria Cretu @ana-mariacretu.bsky.social · 20/01/2026

I am delighted to announce I have joined @cispa.de as a tenure-track faculty earlier this month! I’m really excited to join such a stellar team of security and privacy researchers!

2101

Ana-Maria Cretu @ana-mariacretu.bsky.social · 16/12/2025

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 🧵

2167

Reposted by Ana-Maria Cretu

CISPA Helmholtz Center for Information Security @cispa.de · 06/06/2025

🎧 (German) Podcast Tip: Scammed by an AI voice? If Elon Musk suddenly calls, it's likely a deepfake. Cloning voices with AI takes just seconds — and scammers are all over it. CISPA researcher Lea Schönherr explains how. c’t podcast “They Talk Tech”: 🎙 frauen-technik.podigee.io/30-new-episode

031

Ana-Maria Cretu @ana-mariacretu.bsky.social · 04/06/2025

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!

030

Reposted by Ana-Maria Cretu

strufe @strufe.bsky.social · 09/04/2025

We're organizing the interdisciplinary summer school on privacy again. It focuses on implications of privatizing digital services in the public sector, and it's aimed at PhD students from social sciences, law, and computer science. We emphasize group work, in parallel to lectures: isp.cs.ru.nl/2025/

isp.cs.ru.nl

The 6th Interdisciplinary Summerschool on Privacy (ISP 2025)

The official website of the ISP Summerschool. Find out more about dates, tutorials, talks, the programme, the venue, and travelling suggestions.

021

Ana-Maria Cretu @ana-mariacretu.bsky.social · 28/03/2025

I had the pleasure to give an interview about synthetic data www.wired.com/story/nvidia... by @laurengoode.bsky.social for Wired. I'd like to add a caveat: the promises of synthetic data to solve privacy and data scarcity issues are not yet backed up by scientific evidence.

wired.com

Nvidia Bets Big on Synthetic Data

Nvidia has acquired synthetic data startup Gretel to bolster the AI training data used by the chip maker's customers and developers.

1136

Reposted by Ana-Maria Cretu

Florimond Houssiau @fhoussiau.bsky.social · 26/03/2025

Reading synthetic data papers in healthcare journals is often infuriating as a privacy (ex-)researcher, but I stumbled across one that is really egregious. The authors cite Stadler et al. (the initial work on membership inference attacks on SDG), to (wrongly!) dismiss a SOTA notion of privacy,

A screenshot of the paper, with the following section highlighted:

However, differential privacy has been shown to not fully mitigate the risk of re-identiﬁcation. Stadler and colleagues have shown that under certain circumstances, neither the original implementation of PrivBayes nor PATEGAN (Private Aggregation of Teacher Ensembles Generative Adversarial Network) reliably prevents linkage attacks, leaving some samples vulnerable to membership inference attacks.

111

Ana-Maria Cretu @ana-mariacretu.bsky.social · 19/03/2025

I totally agree: "as a society need to realize that a huge portion of what is being done with generative AI tools is for the type of thing I have described in this article, or for nonconsensual, AI-generated intimate imagery." Must read article: www.404media.co/ai-slop-is-a...

404media.co

AI Slop Is a Brute Force Attack on the Algorithms That Control Reality

Generative AI spammers are brute forcing the internet, and it is working.

010

Reposted by Ana-Maria Cretu

Jaap-Henk Hoepman 🟥 ⬜️ 🟩 ⬛️ @xotoxot.eurosky.social · 06/03/2025

It's my great pleasure to announce that the sixth Interdisciplinary Summerschool on Privacy (ISP 2025) will take place July 6 - 11, 2025, at hotel Erica, Berg en Dal, the Netherlands. This year's topic: Digitalization of the Public Sector: Risks, Rights, and Resistance. Registration is now open!

122

Reposted by Ana-Maria Cretu

Dr Abeba Birhane @abeba.blacksky.app · 18/02/2025

I was in Paris last week for the #AIActionSummit & was honoured to participate in the “AI in public interest” panel. 👇🏾 my thoughts & reflections on what AI in public interest is/isn’t & some concrete steps/initiatives for `bending the arc of AI towards the public interest' aial.ie/pages/aiparis/

aial.ie

Bending the arc of AI towards the public interest

By Abeba Birhane, 18/02/2025 Following the first in Bletchley Park in 2023 and the second in Seoul in 2024, the third AI Action Summit took place in February 2025 in Paris. In the context of previous ...

921779

Ana-Maria Cretu @ana-mariacretu.bsky.social · 10/02/2025

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

122

Reposted by Ana-Maria Cretu

Florimond Houssiau @fhoussiau.bsky.social · 03/02/2025

This paper ieeexplore.ieee.org/abstract/doc..., published at ICAIC25, blatantly plagiarises a paper by myself and colleagues at the Turing,, claiming that they developed the sqlsynthgen package. Although the text has been rephrased, the contents are nearly identical.

123

Reposted by Ana-Maria Cretu

Andrea Ganna @andganna.bsky.social · 28/12/2024

Fantastic, very practical review, on the challenges in data and model anonymization. Also highlights the enormous gap between what is theoretically possible and what used in practice (eg. N>5 rule) www.science.org/doi/10.1126/...

science.org

Anonymization: The imperfect science of using data while preserving privacy

Safe anonymization and sharing of personal data will be achieved by combining formal privacy methods with red-teaming.

0146

Ana-Maria Cretu @ana-mariacretu.bsky.social · 09/01/2025

For my first post on Bluesky, here is a recommended read for anyone interested in the science of re-identification. This is a paper that's elegant, deep, and solves a real problem. I still remember how my mind was blown away when I read it. Congratulations @rocher.lc @yvesalexandre.bsky.social

161