SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Fluidstack is building civilization-scale AI compute infrastructure, acquiring power, designing and operating data centers at scale. The company's mission is to deploy frontier compute faster than anyone else, ensuring AI expands human freedom rather than constrains it.
As Security Architect for AI Infrastructure, you will own the security architecture for the most valuable compute infrastructure in the world—where model weights and frontier AI training happen. This is a rare opportunity to build a security program from scratch, end-to-end, as the company scales across continents.
Key responsibilities:
- Own the unique attack surface of frontier compute: model weights custody, training environment isolation, accelerator and interconnect security, and multi-tenant boundaries under adversarial pressure.
- Design reference architectures and security standards that engineering teams build to, maintaining speed without sacrificing defensibility.
- Architect environments that remain secure even when the workload itself could be adversarial.
- Run security design reviews for systems touching training and inference, from bare metal and fabric through scheduler and storage.
- Stay ahead of the frontier: track emerging accelerators, interconnect generations, confidential computing, and agentic workloads, with architecture ready before deployment.
- Defend design decisions on merit and adapt when closer technical expertise proves a better approach.
The company operates with radical autonomy ("be a barrel"), insane urgency, first-principles reasoning, and a focus on work that genuinely matters. You'll work alongside hardware and software teams spanning data center design, power acquisition, and operations.
REQUIREMENTS:
- Proven experience securing AI, ML, or HPC infrastructure where assets were genuinely high-value targets (lab infrastructure, supercomputing centers, or research environments drawing nation-state attention).
- Deep ability to architect isolation at every layer: hardware, firmware, hypervisor, kernel, container, network fabric, and physical security.
- Systems-level understanding of AI compute infrastructure (accelerators, high-speed interconnects, parallel storage, schedulers) and knowledge of where each component breaks classical security assumptions.
- Original work in systems or ML security—published or shipped—with ability to translate research into architecture engineers actually build.
- Willingness to defend design decisions on merit and change them when proven wrong by deeper technical expertise.
- Bonus: confidential computing and attestation expertise, weights or cryptographic key custody at scale, or publication record in systems/ML security.
