Status: PUBLIC v0.1 (Council-Reviewed)
Note: This document is a public overview. It is not the SGC standard itself.
SGC defines governance requirements for agentic systems — how actions are authorized, how evidence is produced and preserved, how disclosure is controlled, and how systems can be audited without exposing protected internals.
It does not define model architectures, training methods, or claim safety outcomes.
As AI systems become more agentic, they increasingly:
Yet governance practices are often informal, unverifiable, or applied only after failures occur.
SGC addresses this gap by focusing on executable governance controls — governance mechanisms that can be tested, evidenced, and independently reviewed, rather than asserted by policy alone.
SGC is a governance and auditability standard, not a performance benchmark.
Systems may claim conformance only when supported by evidence and appropriate disclosure status.
Public conformance statements are governed and may differ from internal conformance levels.
SGC is currently maintained as an internally finalized standard.
Public review and release occur only through controlled processes.
Governance failures disproportionately affect users and communities with the least ability to contest or remediate harm.
By making governance testable and reviewable, SGC aims to reduce silent failure modes and improve accountability without forcing disclosure of protected system internals.
Built 2025-12-17 22:14 UTC. AICT is not a regulator and not a certification authority.