scx.ai logo

Strategy Blog

Why Sovereign Security Needs Open Models: SCX.ai Joins the Open Secure AI Alliance

SCX.ai has joined the Open Secure AI Alliance (OSAIA), a Linux Foundation Directed Fund advancing open, inspectable models and secure agent harnesses alongside NVIDIA, Microsoft, CrowdStrike, Red Hat, and Cisco. CEO David Keane on why sovereign security needs open models.

By David Keane, Founder and CEO, SCX.ai6 min read

The Black-Box Safety Myth

For the past two years, enterprise and government leaders have been sold a convenient myth about artificial intelligence: that closed, proprietary "black-box" models are inherently safer than open-weight models.

Then real life intervened.

When cyber defenders recently had to contain a sophisticated intrusion at Hugging Face, the opaque safety filters inside commercial closed AI tools blocked their security engineers from running urgent forensic analysis. Unable to tell the difference between an attacker and a defender, the closed tools locked up right when speed mattered most.

How did the security team contain the breach? They ran an open-weight frontier model (GLM 5.2) on their own infrastructure, analysing more than 17,000 actions with full visibility and control.

At SCX.ai, we have made this exact argument from day one: you cannot outsource your operational resilience to a black box you are not allowed to inspect, hosted in a jurisdiction you do not control.

Today, we are proud to announce that SCX.ai has officially joined the Open Secure AI Alliance (OSAIA), a Directed Fund of The Linux Foundation launched alongside NVIDIA, Microsoft, CrowdStrike, Red Hat, Cisco, Palantir, Hugging Face, and more than 120 global technology and cybersecurity leaders.

SCX.ai joins the Open Secure AI Alliance (OSAIA), a Linux Foundation Directed Fund

What Is the Open Secure AI Alliance?

The Open Secure AI Alliance is a global coalition built on a foundational truth: just as open-source software became the bedrock of modern cloud and cybersecurity infrastructure, open and inspectable AI is essential to defending enterprises and nations in the age of autonomous agents.

Operating under the governance of The Linux Foundation, OSAIA brings together frontier model builders, cloud infrastructure providers, and cybersecurity leaders across four core pillars:

  • Open Models: Supporting transparent, inspectable open-weight models that cyber defenders and regulated organisations can audit, customise, and run on their own infrastructure.
  • Open Harnesses & Guardrails: Developing shared frameworks, sandboxes, and identity controls to govern what autonomous AI agents can see, access, and execute in production.
  • Open Skills: Establishing cryptographically verified, vulnerability-tested agent capabilities that protect organisations against prompt injection and tool poisoning.
  • Open Research & Collective Defense (SAFE): Launching the Shared AI Findings Exchange (SAFE) Working Group to turn agentic AI security incidents and near-misses into shared, evidence-based defences for the entire ecosystem.

Why This Matters for Sovereign AI in Australia

At SCX.ai, we define true Sovereign AI as a three-legged stool: (1) onshore Australian data location, (2) local ownership and immunity from foreign jurisdictional reach, and (3) genuine control over the models and secure enclaves running your workloads.

The world needs both proprietary models and open-weight models. Closed frontier models are like Ferraris: high-performance, specialised, and priced accordingly. Open-weight frontier models are the workhorses of enterprise scale: transparent, customisable, dramatically more cost-effective at high token volumes, and rapidly closing the capability gap.

When you deploy autonomous AI agents across banking, telecommunications, health, critical infrastructure, or defence, an AI agent is no longer just a chatbot. It is a system with permissions, tool access, and execution authority. Securing that system requires more than a vendor promise. It requires inspectable weights, verifiable agent harnesses, strict identity boundaries, and sovereign infrastructure where sensitive telemetry never leaves Australian soil.

How SCX.ai Is Putting OSAIA Principles into Practice

As a General Member of the Open Secure AI Alliance, SCX.ai is bringing global open-security standards directly to Australian enterprises and government agencies:

  1. Sovereign hosting of frontier open-weight models. Through the SCX.ai platform, Australian organisations can run leading open-weight models (including Llama-4-Maverick, Qwen3-32B, gpt-oss-120b, gemma-4-31B-it, DeepSeek-V3.1/V3.2, and MiniMax-M2.7) alongside managed access to frontier models like GLM 5.3 and SCX.ai's own sovereign models, MAGPiE and SCX-coder.
  2. Hardened agentic harnesses and enclaves. We are aligning our runtime guardrails, model evaluation pipelines, and secure enclave architecture with OSAIA reference implementations and verification tooling.
  3. Participating in the SAFE Working Group. By contributing to The Linux Foundation's Shared AI Findings Exchange (SAFE), we ensure our customers benefit from real-time, global threat intelligence and hardened operating controls for autonomous agents.

The Future of AI Security Is Open, Inspectable, and Sovereign

Security has never been achieved through obscurity. The safest path forward for Australian business and government is one where defenders have the transparency, tools, and sovereign infrastructure to verify every layer of their AI stack.

We are excited to collaborate with NVIDIA, The Linux Foundation, and our fellow Open Secure AI Alliance members to build a safer, more resilient future for enterprise and sovereign AI.

To learn more about deploying secure, sovereign open-weight and frontier AI models in Australia, visit scx.ai or get in touch with our team.

Related Topics

Open Secure AI AllianceOSAIALinux Foundationopen-weight modelssovereign AIAI securityagentic AINVIDIASAFEcyber defence
Why Sovereign Security Needs Open Models: SCX.ai Joins the Open Secure AI Alliance