SL5 Standards Lead

Security Level 5•San Francisco, CA

5h•Onsite

About The Position

Security Level 5 is a Bay Area AI security nonprofit working with AI labs and US intelligence agencies to defend frontier models from priority nation-state-level threats. We're a small technical team, we move fast, and we're growing quickly this year. The mission Plausibly the most difficult, urgent, and interesting challenge in AI security engineering this decade (for humans, at least): designing the first datacenter purpose-built to Security Level 5, the standard for defending frontier labs against attackers with nation-state resources and billion-dollar budgets. We wrote v0.1 with input from frontier-lab CISOs and US security and intelligence officials. Now we're designing SL5 architectures for real AI workloads and standing up the reference stack to guide lab buildouts. First SL5 components this year, at-scale build in 2027. The stack has to hold against nation-states, keep researchers productive, stay within what labs can pay, and ship in 2-3 years.

Requirements

Authored, owned, or led a technical standard, control framework, or comparable cross-cutting spec, ideally one real organizations had to implement.
Deep security experience in more than one of: systems/network, physical/personnel, hardware/ML infra, government security frameworks.
Fluent with NIST 800-53 / ICD 705 / IL6-class frameworks.
Write clearly and precisely for technical and policy readers.
Want your work to set the bar the whole industry builds to.

Nice To Haves

Wrote or operated under classified or high-sensitivity frameworks.
Hold a clearance or are clearance-eligible.
Have standing relationships across frontier labs, the IC, or datacenter/chip supply chains.
Owned a standard that had to be both secure and adoptable under real cost and speed pressure.

Responsibilities

Lead the SL5 standard end to end.
Keep it technically coherent and drive it from draft to something labs actually adopt.
Lead versioning, control updates, and cross-stream coherence.
Find and fix flaws in the current version and drive its open questions to resolution.
Extend the mappings to IL6, FedRAMP High, and CNSSI 1253, plus the AI-datacenter and AI-SCIF overlays to ICD 705.
Gather lab, datacenter operator, chip provider, and government feedback, and translate their constraints back into the standard.
Make the required-vs-recommended-vs-deferred calls where security and adoptability conflict.
Set the writing and editorial bar for a document read by both deeply technical and senior policy audiences.