We're excited to announce AI Security Bootcamp (AISB), a 4-week intensive program designed to bring researchers and engineers up to speed on security fundamentals for AI systems. The program will cover cybersecurity fundamentals (cryptography, networks), AI infrastructure security (GPUs, supply chain security), and more novel attacks on ML systems (dataset trojans, model extraction). This program will run in-person from 4th Aug to 29th Aug in London, UK. We will cover all expenses.
Apply here to participate in AISB before EOD AoE, 22nd June 2025.
We are also looking for instructors for parts of the program and staff to help with operations. Apply here.
We are running a 4-week program designed to equip AI safety researchers and engineers with critical security skills. We hope you'll leave the program with a well-practiced security mindset that will help you work on impactful projects and make AI systems more secure.
The curriculum includes exercises designed to help you get hands-on experience with securing AI systems, while building and practicing the security mindset. This includes a mix of pair programming exercises, lectures, reading about public vulnerabilities, and chats with experts.
This program is aimed at people who are at ~the start of their journey into security, and have working knowledge in ML (or are willing to brush up using the MLAB curriculum before the program). We encourage you to apply if you think you'll be a good fit regardless of checking all the boxes - if going on technical deep dives and trying to understand how systems work by peeling the layers of abstraction away excites you, we'd love to hear from you.
The content is divided into roughly three sections - introduction to security, cybersecurity for AI infrastructure, and attacks on modern ML pipelines. The exercises are designed to give you hands-on experience with cybersecurity on both the offensive and defensive sides, as well as train your security muscles so that you are more effective at spotting and patching security holes.
We are also looking for more people to help us run the program:
Apply here to participate before end of the day, 22nd June 2025 (anywhere on earth). We will evaluate applications on a rolling basis, but we aim to get back to everyone by 30th June 2025. Please apply earlier if you'd like a decision sooner, and highlight your deadlines in the application form.
We would like applications from people who:
If you're not sure if you check all of these boxes, but think you'll be a good fit for the program, we encourage you to apply.
We expect to have a mix of students as well as professionals, so you should apply if you fit the background criteria.
You'll start the day with a lecture or a deep dive into a significant vulnerability or exploit. After this, you will spend most of your time pair programming to work on the exercises, and reading relevant material. We will plan to wrap up the exercises by dinnertime, after which we'll have some additional reading material to prepare for the next day.
In the last week, you'll work on a capstone project with a mentor - this can be implementing a security solution, replicating a significant cyberattack (in a controlled environment), or trying to break into systems (again, with explicit permission).
The curriculum and the readings will likely not leave much time, so we would recommend not having other major time commitments during the bootcamp. Feel free to reach out if you'd like to discuss your specific circumstances.
Yes, we would recommend having a good background in deep learning (or going through the content for a program like MLAB or ARENA) and being comfortable with Python. We will also send an exercise to help you brush up on essentials a couple of weeks before the program starts.
No, given we have limited spaces, we would be more excited to have participants who can commit to the entire three weeks of the program.
We will help you with accommodation, travel, food, and visas if you need us to. No, roller-coaster rides are not included.
This program is supported by Open Philanthropy.
