Professor of Applied Math and CS at @ImperialCollege. Head of the AI Security and Privacy Lab. @MIT PhD. Road cyclist 🚴♂️. Opinions are my own.
London, England
Joined April 2008
- Tweets1.7K
- Following294
- Followers4.4K
- Likes3K
Pinned Tweet
🚨 New @icmlconf paper from our group on copyright traps. A method for content owners to detect whether their text was used to pre-train or fine-tune an LLM.
1/10 New paper alert! 🚨We investigate the use of copyright traps to detect if Large Language Models (LLMs) have been trained on specific content. More details in thread or find us at ICML 2024 @icmlconf in Vienna 🇦🇹 next week 🧵👇
Another great piece by the group
🎙️ New preprint: You Are the Trigger: Speaker-Activated Backdoors
We introduce a backdoor attack on speech models where the trigger is a target speaker’s voice. The model behaves normally for everyone else, but switches to attacker-chosen behavior when it hears the target. 1/
1
299
Fantastic work by my PhD students and collaborators
New paper: We deploy Claude Code in an autoresearch loop to discover novel jailbreaking algorithms – and it works. It beats 30+ existing GCG-like attacks (with AutoML hyperparameter tuning)
This is a strong sign that incremental safety and security research can now be automated.
5
639
Looking forward to working with Executive Vice-President @Teresaribera, the DG Technology team, and colleagues at the European Commission over the next three years!
To keep pace with fast-evolving digital markets, DG Competition is strengthening its technological expertise at the core.
📢Today, we appoint Professor @yvesalexandre as Chief Technology Officer.
Welcome on board Professor and best of luck in your new role❗️
New work from my team on identifying memorized training samples for free (no extra computational cost 💵)!
New paper accepted @ USENIX Security 2025!
We show how to identify training samples most vulnerable to membership inference attacks - FOR FREE, using artifacts naturally available during training! No shadow models needed.
Learn more: computationalprivacy.github.…
Thread below 🧵
Have you ever uploaded a PDF 📄 to ChatGPT 🤖 and asked for a summary? There is a chance the model followed hidden instructions inside the file instead of your prompt 😈
A thread 🧵
🚨One (more!) fully-funded PhD position in our group at Imperial College London – Privacy & Machine Learning 🔐🤖 starting Oct 2025
Plz RT 🔄
How do you know your synthetic data is anonymous 🥸?
If your answer is “we checked Distance to Closest Record (DCR),” then… we might have bad news for you.
Our latest work shows DCR and other proxy metrics to be inadequate measures of the privacy risk of synthetic data.
Yes yes I know the fundamental law of information recovery and differential privacy, but if there are really just a few summary statistics, surely it should be anonymous? 🥸
I definitely used to think this, until we started looking into it two years ago.
A thread 🧵
Huge congrats my students and collaborators for winning the Best Paper Award at @satml_conf! 🎉
And we were lucky enough to win the @satml_conf best paper award!!🌟
@_igorshilov @yvesalexandre @MarekRei @shubhamjain0594 @ManuelFaysse
🚨 Breaking: UK former Secretary of State for Defence discovers good-guys-only backdoor pesky cryptographers have been hiding from governments for years! 🤣🤣🤣
Just to be clear: this has nothing to do with allowing authorities access or not. If you have the key 🔑 you can decrypt.
There are a range of views on online safety, see e.g. the debate with @tomhfh, but this is nonsense technically.
1
1
210
Glad to see our work on the privacy of synthetic data references in the International AI Safety Report!
Paper: arxiv.org/abs/2306.10308
Work done with @matthieu_meeus, Florent Guépin, and @AnaMariaCretu5 at @imperialcollege.
Today, we are publishing the first-ever International AI Safety Report, backed by 30 countries and the OECD, UN, and EU.
It summarises the state of the science on AI capabilities and risks, and how to mitigate those risks. 🧵
Link to full Report: assets.publishing.service.go…
1/16
🚨 In a new paper in @NatureComms, we propose a scaling law for identification technologies, from browser and device fingerprinting 🌐 to facial recognition 📸 and stylometry ✍️.
A thread 🧵:
Join us at Imperial College for an exciting event on the future of privacy in machine learning! 🔒🤖The application for lightning talks is open.
🗓️Date: Feb 4 @ 6pm
📍Imperial College London
We're very excited to host this meetup and we'd be thrilled to see you there!
imperial.ac.uk/events/183182…
And a huge shoutout to my co-organizers @matthieu_meeus, Nataša Krčo (not on twitter), @yvesalexandre and @ImperialDSI team!
The FT said this is where the cool kids hang out. So, here I am, trying to be cool. bsky.app/profile/yvesalexand… 🐑👴
This quoted post is unavailable.
1
326
Great piece by Melissa on Nightshade, what it provides and what it doesn't do.
Meet @TheGlazeProject, the AI lab behind Glaze and Nightshade, two prominent weapons in an artist’s arsenal against nonconsensual AI scraping. The lab wants to tilt the balance of power from Big Tech back to individual creators. But are their tools enough?technologyreview.com/2024/11…
2
411
Short piece by @imperialcollege on our recent award 🏅at @acm_ccs for QueryCheetah 🐆, a tool for fast automated discovery of vulnerabilities in data APIs and query-based systems imperial.ac.uk/news/257836/d…
Interested to work in a fast-paced research environment on the intersection of privacy and ML/AI (LLMs, diffusion models, and beyond)?
Come join us 👩🎓👨🎓! Our research group at @imperialcollege has openings for fully-funded PhD students (Nov 15 deadline).
Plz RT 🔁
Recently, we used canaries 🐦 to study LLM memorization (arxiv.org/pdf/2402.09363), developed methods to efficiently discover 🔎 vulnerabilities in QBSs (arxiv.org/abs/2409.01992), to perform ZK MIA against aggregate location data (arxiv.org/abs/2406.18671).
1
73
Great piece in @Nature on @GoogleDeepMind's new watermarking technique for LLMs output 🤖 (with comments from yours truly)
nature.com/articles/d41586-0…
🧵Quick summary:
- the paper improves previous work with tournaments 🏅 making it easier to detect the watermark (@tomgoldsteincs)
Our new work on fast automated discovery of vulnerabilities in query-based systems has just received a distinguished paper award at @acm_ccs 🎉🎊
New paper alert! 🚨 Introducing QueryCheetah 🐆 – our novel method for fast automated discovery of privacy attacks against query-based systems and a recipient of the Distinguished Paper Award 🏆 at @acm_ccs. arxiv.org/abs/2409.01992 🧵(1/n)
13
631