FASR-HARDWARE-2026 // COORD-HERO
Preventing risks from advanced AI requires international coordination.
Preventing risks from advanced AI requires robust technical building blocks. This 8-week, fully funded track within the Frontier AI Security Residency is engineered for engineers and researchers building the hardware mechanisms that let states verify each other's compliance with agreements on frontier AI development.
Our focus on this track: build the on-chip and off-chip mechanisms, like compute accounting, tamper-evident sensors, hardware-backed attestation, that turn claims about AI development into verifiable outputs.
FASR-HARDWARE-01 // VERIFICATION-BASE
International Trust Requires Robust Verification.
Rival states have cooperated on catastrophic, shared risks before, but doing so is easier when each side can confirm the other is holding to its word. Hardware-enabled verification is what makes that confirmation possible for AI.
As AI grows increasingly capable, leading nations must come together to cooperate on reducing catastrophic risks posed by the most advanced AI systems. Doing so will require developing the hardware technology that can enable mutual trust between nations about their AI workloads.
FASR-HARDWARE-02 // ADVISORY-TRACKS
Work with Expert Mentors to verify AI compute.
Residents are paired with mentors from hardware security, cryptography, and AI governance who work on building these mechanisms.
Applied Track: Embed yourself within an organization currently working on building AI compute verification technology and assist your team with their ongoing projects. Ideal for those with extensive engineering backgrounds who are comfortable getting up to speed on projects in a short amount of time.
Research Track: Contribute to the nascent research field of AI compute verification with guidance from experienced mentors. Take ownership of a research project that will allow you to be a leading expert on that particular topic. Ideal for those with academic and research backgrounds who can think holistically about a particular research agenda.
FASR-HARDWARE-03 // FIELD-TARGETS
Target specific focus areas.
A non-exhaustive list of potential projects hardware residents will work on. Feel free to propose your own customized exploration vectors!
Cryptographic Verification
Extend zero-knowledge proofs to neural networks for a cryptographic proof a model ran as declared, without exposing the underlying model weights.
Anti-Tamper Systems
Build and empirically validate options, like tamper-responsive enclosures, side-channel shielding, and tamper-evident security cameras, to protect AI hardware and hardware-enabled verification mechanisms from physical tampering.
Compliance Infrastructure
Build a hardware-enabled auditing architecture that reliably and efficiently logs low-level GPU activity.
External Monitoring
Utilize side-channel sensors, like power draw, electromagnetic emissions, and network taps, that verify off-chip what a cluster actually computed, independent of anything the chip itself reports.