The MLSec Laboratory is a research branch of the PRALab, at the University of Cagliari. We investigate topics at the intersection of ML and computer security.
Cagliari, Italy
Joined December 2021
- Tweets156
- Following472
- Followers894
- Likes78
Machine Learning Security Laboratory retweeted
What is refusal ablation actually doing inside a language model?
Our new preprint shows that standard refusal ablation can be interpreted as a minimum-confidence latent-space evasion: it projects the representations onto the boundary of a refusal probe.
Can we push beyond it?
Machine Learning Security Laboratory retweeted
We are excited to present a new event of our seminar series on ML Security!
We will host @chwress (KIT) on November 26th, 2025, at 5 pm CET.
Free registration at the link below:
events.teams.microsoft.com/e…
Machine Learning Security Laboratory retweeted
The @mlsec_lab online #seminar series continues! On November 26th, you can join @chwress for his insights on “Explanation-Aware Attacks and the Limits of using #XAI in #ComputerSecurity”.
If we were you, we wouldn't miss it.😉
Registration: events.teams.microsoft.com/e…
#ExplainableAI
Machine Learning Security Laboratory retweeted
Language models encode refusal as a manifold expressed through multiple, closely related directions.
In our latest #AAAI26 paper, with @RaffaeleMura3, we bridge LLM safety and interpretability by answering the question:
👉 Do LLMs encode refusal behavior as a manifold?
Machine Learning Security Laboratory retweeted
We are excited to present a new event of our seminar series on ML Security! We will host @XinCynthiaChen (ETH Zurich) on October 22nd, 2025, at 5 pm CEST.
Free registration at the link below:
events.teams.microsoft.com/e…
Machine Learning Security Laboratory retweeted
October 22nd: The @mlsec_lab seminar series continues!
@XinCynthiaChen from the @ETH_en will share her knowledge on “Learning Safety Constraints for Large Language Models”.
Register for the online seminar today: elsa-ai.eu/mlsec-seminar-ser…
Machine Learning Security Laboratory retweeted
Perfect for the weekend: @mlsec_lab's seminar series "UNVEILING THE FUTURE OF MACHINE LEARNING SECURITY"
It's an amazing resource to dive into the world of cutting-edge machine learning security. 🛡️ Enjoy!
elsa-ai.eu/mlsec-seminar-ser…
#machine_learning #security #seminar #ELSAAI
Machine Learning Security Laboratory retweeted
Delve into the future of Machine Learning Security with seminars from @mlsec_Lab. World-class experts, past seminars, and free courses on YouTube. Open to all! elsa-ai.eu/mlsec_seminar_ser…
Machine Learning Security Laboratory retweeted
Great news! 📢
The CFP of the 16th ACM Workshop on Artificial Intelligence and Security (AISec 2023) is out!
Co-located with @acm_ccs
Deadline: June 24, 2023
Website: aisec.cc
w/ @florian_tramer @xinyun_chen_
@adversarial_ML @trustworthy_ml @RedTeamVillage_
Machine Learning Security Laboratory retweeted
Happy to share that I have been promoted to Associate Professor @univca. Thanks to all colleagues and collaborators who made it possible. Looking forward to the next steps!
Machine Learning Security Laboratory retweeted
We just released a new preprint version of "Indicators of Attack Failure: Debugging and Improving Optimization of Adversarial Examples". A lot has changed, thanks to reviewers and extraordinary teamwork. Check it out!
BTW accepted at NeurIPS 2022 :)
arxiv.org/abs/2106.09947
Machine Learning Security Laboratory retweeted
📢We are proud to announce that our paper "Indicators of Attack Failure: Debugging and Improving Optimization of Adversarial Examples " has been accepted to #Neurips2022!
w/ @biggiobattista, @zangobot, @ambrademontis, @sotgiu_angelo, Nicholas Carlini, and Fabio Roli
1/n
Machine Learning Security Laboratory retweeted
Today I will present our work ImageNet-Patch @shifthappens_22 (@icmlconf)!
Workshop: shift-happens-benchmark.gith…
Preprint: arxiv.org/abs/2203.04412
Code: github.com/pralab/imagenet-p…
Benchmark: github.com/shift-happens-ben…
w/@biggiobattista @DAngioni97 @ambrademontis @zangobot @sotgiu_angelo
Machine Learning Security Laboratory retweeted
Surprise! Blaine will be joined by some special guests today: co-authors @biggiobattista and Pavel.
The band is back together!
Join for insights into the foundations of modern ML poisoning, stay for the co-author banter.
Drop by @robusthq in a couple hours! (Reg. below)
For some backstory on Blaine and the work for which @biggiobattista, he and Pavel (at @uni_li) received @icmlconf's Test of Time Award, join us for a Friday morning (Pacific time) chat! ↓↓↓
What questions should I ask? 🤔
Machine Learning Security Laboratory retweeted
Ever wondered which ML Security threats occur in the wild, what determines exposure to an attack, or what and why is perceived by practitioners as a threat for ML? We answer these questions in our survey study with 139 participants from organizations around the world! A 🧵 1/11
Machine Learning Security Laboratory retweeted
For now, i am really honored and proud to have contributed to a @dagstuhl seminar! I am looking forward for 2 1/2 more exciting days!
Machine Learning Security Laboratory retweeted
Really happy to be here at @dagstuhl with @maurapintor and @biggiobattista presenting our recent research on the aspects we think are broken in #poisoning and #evasion literature.
#MLsec #AI #adversarial #security #DataPoisoning @adversarial_ML @aivillage_dc @RedTeamVillage_
Machine Learning Security Laboratory retweeted
Where ML Security is Broken and How to Fix It
Excited to share some thoughts at @dagstuhl - Seminar on Security of Machine Learning
w/ @cinofix @biggiobattista
