Research Engineer / Research Scientist, Remote Compute Accounting in London

The team is focused on answering a crucial question: can we verify what is happening inside AI datacenters? This could unlock international cooperation, help countries protect their sovereignty, and enable trustworthy adoption of AI in high-stakes industries. SASH’s Verification team builds and tests tools to verify agreements about AI, prototypes new verification mechanisms, leads international research collaborations, and works with policymakers worldwide.

The hardest question in verification is whether a datacenter is running something else alongside the declared model. We are interested in software-only approaches to this problem. If a verifier can keep a cluster's accelerators demonstrably occupied, then there is no spare capacity for an undeclared training run. The time a cluster takes to answer is evidence about what hardware it has and what that hardware is doing.

Your work would involve owning the question of how much of a cluster can be accounted for from outside it, designing challenge protocols, measuring whether real accelerators can satisfy them, microbenchmarking, writing challenge kernels, and characterising high-bandwidth memory under adversarial conditions. You'd work alongside cryptographers and hardware-security colleagues, and you'd be expected to publish findings.

Representative projects include:

• Implementing a candidate challenge function and characterising honest-response latency across real cluster topologies.
• Establishing how much spare capacity a cluster could retain while still passing every challenge.
• Running an adversarial compression study.
• Designing and evaluating post-training that raises the cost of compressing weights.
• Quantifying how much memory a prover could plausibly hide.
• Extending the approach from memory occupancy to compute.
• Taking the protocol end to end against a live inference cluster.
• Designing a red-team/blue-team competition for the protocol.

About you: We care more about ability and trajectory than years of experience. You may be a good fit if you:

• Have GPU performance or systems measurement experience or a cryptographic background.
• Are rigorous about measurement.
• Can hold an adversary in your head while optimising something.
• Are comfortable working on a protocol that may not work.
• Thrive in ambiguous, early-stage environments.

Strong candidates may also have:

• CUDA or equivalent low-level accelerator programming experience.
• Compression, information theory, or model quantisation experience.
• Background in timing-based protocols.
• Experience characterising undocumented hardware.
• Familiarity with LLM inference internals.

Role Logistics & Benefits: This is a remote role that can be based in Singapore, the UK, or remotely. For candidates interested in relocating, SASH can provide visa sponsorship in the UK. Our team is globally distributed, so remote team members should maintain some working-hour overlap with colleagues across regions.

Compensation takes location, experience, and level into account. Salaries above the stated range may be available for exceptional candidates. Benefits include competitive benefits and leave policies.