Overview
The AISF DC Reading Group is a technical forum for practitioners and researchers focused on the field of AI Security. We examine the engineering paradigms, hardware constraints, and defensive architectures required to secure frontier models.
Our goal is to build the situational awareness necessary to navigate the intersection of systems safety, hardware security, and emerging AI threat models. We prioritize rigorous exchange over general overview, deep-diving into the foundational texts and research that underpin a secure AI ecosystem.
Logistics
- Held monthly-ish in Washington DC
- For invite to in-person reading groups or access to the online reading bank and forum space, please register your interest below:
If you are interested in joining the reading group, please fill out this registration of interest form.
This reading group will:
- Provide deep dives into foundational technical literature and cutting-edge research in the AI security field.
- Provide an advanced community of technical experts, policymakers and national security professionals to discuss and deliberate over developments in the AI security space
- Facilitate connections between participants and AI security organizations.
- Evaluate the feasibility of hardware governance, model weight protection and infrastructure hardening against specific threat models.
Example Curation
- Flexheg/Chipgov papers
- "Defensive Acceleration" strategies using AI to harden cloud infrastructure
- Nancy Leveson, Engineering a Safer World (selected chapters)
- Ross Anderson, Security Engineering (selected chapters)
- Google's Secure and Reliable Systems (selected chapters)
- AI Control technical papers
- Cyberoffensive capability elicitation
- Steganography: attacks and mitigations
- TEEs and Confidential Inference
- SL5 task force outputs
Am I a good fit for this workshop?
We are looking for reliable, high-engagement contributors who are committed to deeply understanding the AI security field. We value participants who seek the situational awareness necessary to provide multi-faceted approaches to the space's most pressing open problems.
We are most excited about people with:
- Strong information security backgrounds, especially in hardware engineering or experience in high-stakes settings (e.g. defence, health, finance, critical infrastructure …).
- Technical policy expertise, including but not limited to, compute governance or export controls
- Those without any professional experience within this field, but can prove to have interesting opinions which will push this group forward!
Current Reading Track
Session | Topic | Primary Text |
Session 1 | Insider Risk | Redwood Blog posts on Insider Risk |
Session 2 | Hardware Security | SL5 Taskforce’s Novel Recommendations |
Session 3 | AI Assurance | FlexHEG Reports |
Session 4 | Output correctness | Secure program synthesis (foundational texts and papers) |
For inquiries contact [email protected].