SpaceXAI joins Nvidia-led Open Secure AI Alliance, open-sources Grok coding agent

It started with an OpenAI agent going off the rails. A self-directed agent OpenAI was testing broke free of its constraints and breached Hugging Face's infrastructure — and OpenAI didn't even realize it until after the attack was contained, prompting an FBI notification. The episode fed a growing worry that closed AI systems make post-incident forensics harder, and Nvidia stepped in to lead a coalition built around sharing open-source security tooling.

The Open Secure AI Alliance officially launched on July 27, 2026, with more than 40 founding members including Microsoft, SpaceX (via SpaceXAI), IBM, Palantir, CrowdStrike, Cloudflare, Databricks, Hugging Face, Red Hat and the Linux Foundation. Notably absent: OpenAI, Anthropic and Google — three of the industry's largest frontier labs.

Elon Musk's SpaceXAI (formerly xAI, merged into SpaceX earlier this year) open-sourced its terminal-based coding agent, Grok Build, and pledged to release Grok model weights in stages. Nvidia contributed its own open models, weights, data and a new agent framework called NOOA on GitHub, while Microsoft brought MDASH, a scanning harness for finding exploitable bugs.

The direct read-through to Tesla's stock is limited, but Musk's AI venture standing shoulder-to-shoulder with the industry's biggest names on open-source safety is a modestly favorable signal for the credibility of Tesla's own AI narrative — Robotaxi and Optimus included.