AI705 is an SL5 Task Force project for adapting facility-security standards to frontier AI infrastructure.

About This Document

AI705 is a proposed gap analysis and technical guidance project for applying ICD 705 and the IC Technical Specifications for ICD/ICS 705 to frontier AI datacenters. Its purpose is practical: give labs, datacenter operators, colocations, facilities teams, and accrediting officials a shared reference for physically secure environments that handle frontier model weights.

The output is a technical guidance document that states, for each source requirement, whether it applies as written, applies with AI/datacenter-specific modification, is inapplicable, or leaves a gap requiring new controls. This follows the same source-to-guidance discipline used in the SL5 Standard for AI Security, but the unit of analysis is an ICD 705 or IC Tech Spec requirement rather than a NIST SP 800-53 control.

Why The Threat Vectors Matter

AI705 now treats weight theft, secret theft, and sabotage as separate control outcomes. A facility measure that protects stored checkpoints may not protect construction plans, operational telemetry, credentials, or power and cooling paths from sabotage.

Weight theft still carries three categories: stored weights, training systems, and inference systems. A single facility baseline is useful only if those category differences stay visible while secret theft and sabotage are reviewed as first-class outcomes.

About The Security Level 5 Task Force

The SL5 Task Force is a non-profit cross-industry effort working to ensure frontier AI infrastructure can achieve nation-state-level security by 2028/2029. Founded in March 2025, we are a core team of engineers and security strategists leading a 100-person technical track comprising security engineers from frontier AI labs, government security specialists, and datacenter colocation providers, alongside an executive track of AI industry security leaders providing steering input.

Since its founding, the Task Force has conducted workshops and research programs to clarify what it takes to reach Security Level 5 in a way that is sensitive to competitive pressures, the need to maintain speed of innovation, and the reality of a rapidly shifting threat landscape. Our mission is to create the optionality for frontier AI labs to reach Security Level 5 in the coming years, and to be able to activate that security level within six months of choosing to do so.

In service of that mission, we convene technical work that clarifies what needs to be done early to preserve that optionality. AI705 is one output of that effort, focused on the physical, construction, accreditation, and side-channel questions raised by applying ICD 705 to frontier AI facilities.