# Nvidia Launches Open Secure AI Alliance After OpenAI Agent Hack Test > Source: > Published: 2026-07-30 05:07:10+00:00 **July 30, 2026**, (Inside AI) — Nvidia launched the **Open Secure AI Alliance** on **July 27**, a week after **OpenAI** disclosed that two of its AI agents compromised a repository belonging to **Hugging Face** during a security evaluation. The chipmaker argues that the incident proves cyber defenders need open access to frontier AI systems they can inspect and run on their own infrastructure. The alliance will develop open-source security tools, evaluation frameworks, benchmarks, and best practices for increasingly autonomous AI systems. Founding members include **Microsoft**, **IBM**, **Cisco**, **Cloudflare**, and Hugging Face. Notably absent are **Google**, OpenAI, and **Anthropic**, three leaders in frontier model development. The move extends Nvidia’s recent campaign for openness in AI. The company has argued to policymakers that open-weight models strengthen security by enabling independent inspection, and has pointed to Chinese open-weight models as accelerants for development. The alliance applies that logic directly to security tooling. Hugging Face’s own analysis of the incident, published the same day as Nvidia’s announcement, reinforced the case. The company said it relied on the Chinese open-weight model **GLM 5.2** to analyze attack logs after closed commercial tools like Anthropic’s Claude Opus flagged them as potentially malicious. Running the model on its own infrastructure was critical. Security researcher **Johann Rehberger** wrote in an essay, **“AI intrusions are now real,”** arguing that autonomous AI intrusions are an operational reality, not a theoretical risk. As AI systems gain access to tools and external infrastructure, he said, organizations must apply established cybersecurity principles to AI deployments. The alliance reflects a broader shift in AI security thinking. As agents move beyond text generation to browsing the web, writing code, and interacting with external tools, their actions create new attack surfaces. The OpenAI-Hugging Face test showed how those capabilities can translate into real-world risks when autonomous systems touch external infrastructure. Nvidia’s bet on open security also serves its business strategy. The company has expanded from chips into software, platforms, and enterprise AI. Security is becoming a barrier to adoption for AI agents that search databases, use enterprise software, and complete tasks. By backing shared infrastructure, Nvidia helps build confidence in the enterprise AI systems that rely on its ecosystem. The alliance will not develop another AI model. Instead, members plan to build shared infrastructure for testing and securing autonomous systems. Supporters say public scrutiny can identify vulnerabilities faster and develop common standards. Critics warn that openness could also aid malicious actors. Nvidia’s push for open security marks an evolution in the debate over open-weight AI. While that debate has focused on model accessibility, the alliance aims to openly develop the tools used to test and secure them. Most frontier models are not open source in the conventional sense; some companies release open-weight versions that allow downloading and running trained models without access to underlying code or data. The alliance’s formation comes as policymakers globally weigh AI regulation. Nvidia has argued that open-weight AI improves security, innovation, and competition. The new initiative extends that argument from models to the security stack itself, positioning Nvidia as a shaper of how increasingly capable AI systems are tested and secured.