News
-
24 September 2026
|Celebrating three papers accepted to NeurIPS'26! Auditing Privacy Leakage in Tabular Foundation Model Embeddings studies sensitive attribute recovery, RADIUM: RadioActive Decay of Image-Underlaid Marks investigates watermark persistence across model training, and Benchmarking Membership Privacy Risks in Preference-Based LLM Post-Training advances privacy audits for preference-trained LLMs. Congratulations to all the authors!
-
21 August 2026
|Our paper AgentProv: Auditing Agentic LLM API Providers via Tool-use Policy Probes has been accepted to EMNLP'26! AgentProv audits whether an agentic LLM API serves its claimed model by fingerprinting its tool-selection distribution. Congratulations to all the authors!
-
20 June 2026
|Excited to announce two papers accepted to ECCV'26! MGI: Member vs Generated Inference distinguishes training members from model-generated images using Data Circuit Breaker, while MultiMem: Measuring and Mitigating Memorization in Multi-Modal Contrastive Learning introduces a framework to measure and reduce memorization across modalities. Congratulations to all the authors!
-
30 April 2026
|Celebrating two papers accepted to ICML'26! Finding DoRI: Discovery of Retained Images in Diffusion Models challenges the assumption that memorization is local, while Concept Removal for Frontier Image Generative Models introduces persistent concept removal through interventions in the model backbone. Congratulations to all the authors!
-
20 April 2026
|Our most successful ICLR ever — 5 accepted papers at ICLR'26! Natural identifiers for privacy and data audits in LLMs, data provenance for image auto-regressive generation, curation leaks: membership inference attacks against data curation, SERUM: simple, efficient, robust, and unifying marking for diffusion-based image generation, and benchmarking empirical privacy protection for LLM adaptations (ORAL)!
-
09 November 2025
|Excited to announce three papers accepted to AAAI'26! Our work on mitigating unsafe text in image generative models, demystifying foreground and background memorization in diffusion models, and our graph stealing paper which received an ORAL presentation!
-
18 September 2025
|Excited to share three papers accepted to NeurIPS'25 on watermarking bitwise autoregressive models (BitMark), memorization in GNNs, and strong membership inference attacks on massive datasets and large language models!
-
01 May 2025
|Thrilled to announce three papers accepted to ICML'25 on privacy attacks on image autoregressive models, dataset inference with synthetic data, and efficient and private soft prompt transfer for LLMs!
-
26 February 2025
|Big congrats on our paper CDI: Copyrighted Data Identification in Diffusion Models accepted to CVPR'25! This is the first method to detect if a given dataset was used during the training of diffusion models.
-
22 January 2025
|SprintML has three accepted papers for ICLR'25! What a successful start of the year.
-
09 December 2024
|Our work on Private Prototypes for efficient and private learning on unbalanced datasets was accepted to AAAI'25.
-
11 November 2024
|Our work on text editing in diffusion models got the best poster award at ML in PL!
-
01 November 2024
|We are extremely proud on the four papers we got in to NeurIPS'24. Happy to share the blog posts soon.
-
01 June 2024
|We have two new openings for HiWi positions!
-
16 January 2024
|Check our new paper accepted at ICLR 2024 on memorization in self-supervised learning!
-
29 October 2023
|The best poster award at MLinPL for our work on B4B (Bucks for Buckets) Active Defenses against Stealing Self-Supervised Models.
-
21 September 2023
|Check our new paper accepted at NeurIPS 2023 on active defenses against stealing of SSL encoders!
-
21 September 2023
|Check our new paper accepted at NeurIPS 2023 on P2P collaborative learning secure against malicious servers and robust to malicious clients!
-
21 September 2023
|Check our new paper accepted at NeurIPS 2023 on individualized DP-SGD!
-
21 September 2023
|Check our new paper accepted at NeurIPS 2023 on privacy for prompts!
-
21 September 2023
|Check our 4 new papers accepted to NeurIPS.
-
25 May 2023
|Only 99 days until the SprintML lab will launch at CISPA.