Application Guide
How to Apply for Research Engineer/Scientist, Remote Compute Accounting
at SASH
🏢 About SASH
SASH is a Singapore-based nonprofit dedicated to building AI safety research capacity and governance expertise across Asia. Unlike typical AI labs, SASH focuses on the intersection of technical AI safety and policy, offering a unique opportunity to work on high-impact, pre-paradigmatic problems with a mission-driven team. Employees value the chance to shape the emerging field of AI governance while working remotely with a diverse, international group.
About This Role
This role involves designing and testing verification protocols to detect undeclared AI model training in datacenters, using hardware measurement and cryptographic approaches. You will microbenchmark GPUs, characterise memory bandwidth and compression feasibility under adversarial conditions, and collaborate with cryptographers and hardware-security experts to construct hardness assumptions. The work is critical for ensuring accountability in AI compute usage and has direct implications for AI safety and governance.
💡 A Day in the Life
A typical day might involve running microbenchmarks on GPUs to measure memory bandwidth under different compression schemes, then analyzing data for anomalies that could indicate undeclared training. You might also meet with cryptographers to refine hardness assumptions or brainstorm new challenge protocols, and document findings for a paper or internal report. The work is highly collaborative and iterative, with a focus on rigorous measurement and adversarial thinking.
🚀 Application Tools
🎯 Who SASH Is Looking For
- Strong background in GPU performance engineering, systems measurement, or applied cryptography, with hands-on experience in microbenchmarking and hardware characterisation.
- Comfortable with adversarial reasoning and designing protocols that remain robust under malicious conditions, including timing and compression side-channels.
- Experience working in ambiguous, early-stage research environments, with a track record of publishing findings, including negative results.
- Familiarity with datacenter hardware, cluster management, and the challenges of verifying compute usage without direct access to internal telemetry.
📝 Tips for Applying to SASH
Highlight any past work on hardware measurement, GPU benchmarking, or cryptographic protocol design—be specific about the tools (e.g., CUDA, NVML, ROCm) and metrics you used.
Emphasize experience with adversarial conditions or security analysis, especially if you've designed protocols to detect cheating or resource misuse.
Discuss your approach to publishing negative results and your willingness to work in ambiguous, early-stage research settings.
Mention any familiarity with AI safety or governance concepts, even if from a technical perspective, to show alignment with SASH's mission.
Tailor your CV to include projects where you characterised memory bandwidth, compression, or timing under constraints, as these are core to the role.
✉️ What to Emphasize in Your Cover Letter
["Your hands-on experience with GPU microbenchmarking and hardware characterisation, including specific examples of protocols you've designed or tested.", 'Your ability to reason adversarially and construct hardness assumptions, ideally with examples from past work in security or verification.', 'Your motivation to contribute to AI safety and governance, particularly in the context of remote compute verification for datacenters.', 'Your comfort with ambiguity and early-stage research, and your track record of collaborating with cross-functional experts like cryptographers and hardware-security specialists.']
Generate Cover Letter →🔍 Research Before Applying
To stand out, make sure you've researched:
- → Read SASH's published reports and blog posts to understand their specific focus areas within AI safety and governance, especially any work on compute verification.
- → Research existing techniques for datacenter compute verification, such as remote attestation, proof-of-learning, and hardware-based measurements, to identify gaps SASH aims to fill.
- → Look into the challenges of detecting undeclared AI training, including recent papers on GPU side-channels and resource accounting.
- → Understand Singapore's role in AI governance and how SASH's nonprofit status might influence its research agenda and partnerships.
💬 Prepare for These Interview Topics
Based on this role, you may be asked about:
⚠️ Common Mistakes to Avoid
- Focusing only on theoretical cryptographic knowledge without demonstrating hands-on hardware measurement experience.
- Ignoring the adversarial aspect of the role—many candidates discuss standard benchmarking but fail to address how protocols hold up under malicious conditions.
- Overlooking the nonprofit and Asia-focused context of SASH, treating it like a typical tech company instead of a mission-driven research organization.
📅 Application Timeline
This position is open until filled. However, we recommend applying as soon as possible as roles at mission-driven organizations tend to fill quickly.
Typical hiring timeline:
Application Review
1-2 weeks
Initial Screening
Phone call or written assessment
Interviews
1-2 rounds, usually virtual
Offer
Congratulations!