📢 New research from Google DeepMind & Google Research! We tackle the challenge of building AI assistants that leverage your data for complex tasks, all while upholding your privacy. 🤖🔐 Dive into our paper for the full details: arxiv.org/pdf/2408.02373 TLDR in 🧵

🚨New Preprint: "Backdooring Bias into Text-to-Image Models" (arxiv.org/pdf/2406.15213) Ever wondered how text-to-image (T2I) models could spread political bias in #Election2024? 💡We introduce a new attack vector by embedding backdoors in T2I models using implicit biases!

🧙 I am recruiting PhD students and postdocs to work together on making sure AI Systems and Agents are built safe and respect privacy (+ other social values). Apply to UMass Amherst Manning College of Information & Computer Sciences and enjoy a beautiful town in Western Massachusetts. Reach out if you have questions!

you cannot deny that the problem with the french language pack will not really bother you after that

OpenAI Operator enables users to automate complex tasks, e.g., travel plans. Services, e.g., Expedia, use chatbots. Soon, these two ends are going to communicate, forming agentic networks. What would these networks enable? what are their risks? and how to secure them? 🧵1/n

How Sudokus can waste your money? If you are using reasoning LLMs with public data, adversaries could pollute it with nonsense (but perfectly safe!) tasks that will slow down reasoning and amplify overheads 💰 (as you pay but not see reasoning tokens) while keeping answers intact

Nerd sniping is probably the coolest description of this phenomena ( Wojciech Zaremba et al described it recently), but in our case overthinking didn't lead to any drastic consequences besides higher costs.

Amazing opportunity to do ground breaking work in LLMs!

(1/n) In our #ICLR2025 paper, we explore a fundamental issue that enables prompt injections: 𝐋𝐋𝐌𝐬’ 𝐢𝐧𝐚𝐛𝐢𝐥𝐢𝐭𝐲 𝐭𝐨 𝐬𝐞𝐩𝐚𝐫𝐚𝐭𝐞 𝐢𝐧𝐬𝐭𝐫𝐮𝐜𝐭𝐢𝐨𝐧𝐬 𝐟𝐫𝐨𝐦 𝐝𝐚𝐭𝐚 𝐢𝐧 𝐭𝐡𝐞𝐢𝐫 𝐢𝐧𝐩𝐮𝐭 ✅ Definition of separation 👉 SEP Benchmark 🔍 LLM evals on SEP

The Privacy Preserving AI workshop is back! And is happening on Monday. I am excited about our program and lineup of invited speakers! I hope to see many of you there: ppai-workshop.github.io

Amazing forward-looking paper on how collaboration could be done where you and I have different perspectives.

Our IEEE S&P SAGAI workshop on systems-oriented security for AI agents has speaker details (abs/bio) on the website now: sites.google.com/ucsd.edu/sagai… We look forward to seeing you in San Francisco on May 15! As a reminder, we are running this "Dagstuhl" style - real discussions.

I am looking for a postdoc to work on multi-agent safety problems, if you are interested or know anyone let me know: forms.gle/NFuYLKj53fVwdW…