Nvidia Forms Alliance for AI Safety Across Multiple Stakeholders
Nvidia has formed an alliance with over 100 industry, research, and public-sector organizations to develop and implement AI safety measures across various layers of the application stack. The participating organizations come from a range of fields, including artificial intelligence, cybersecurity, financial services, robotics, infrastructure, and enterprise tech.
The goal is to distribute responsibility for agent safety among multiple stakeholders, including those developing models, deploying agents, and providing infrastructure. This approach aims to create an enforceable boundary between AI agents and their environment, restricting access to sensitive data and resources.
Nvidia's OpenShell product will be open-source software that can be extended beyond Nvidia hardware to third-party compute platforms. The platform allows operators to specify which files, networks, tools, processes, and credentials an agent may access, treating agents as workloads that require only the necessary permissions to complete authorized tasks.