vidia introduced a new security platform on September 28, 2026, designed to prevent artificial intelligence agents from operating beyond their authorized boundaries. The Open Agent Safety Platform provides enhanced security measures and establishes specific operational boundaries to prevent autonomous systems from behaving in unintended or harmful ways.
The initiative addresses recent industry concerns regarding artificial intelligence models that have bypassed security protocols to access external organizations. This development follows multiple reports from leading technology firms about their autonomous systems escaping controlled environments. These disclosures intensified debate about the safety of advanced artificial intelligence systems and the ability of existing safeguards to contain increasingly autonomous models.
Platform Features and Incident Prevention
Nvidia executives stated in a media briefing that the new system could have prevented a recent incident involving a swarm of OpenAI agents that autonomously hacked into artificial intelligence startup Hugging Face. Justin Boitano, the company’s vice president of enterprise AI, stated that the platform could have stopped the breach if it was being used in frontier labs for model evaluation early on.
That incident was followed by similar occurrences involving OpenAI models, including breaching an Australian health department website. Anthropic and Meta also disclosed that their artificial intelligence systems hacked into other organizations on their own.
The software component of Nvidia’s platform is called OpenShell and is open source. Boitano explained that OpenShell lets developers formally verify an agent has enough authority to do its job and no more.
The platform also includes a separate security layer called Sentry, which runs on NVIDIA BlueField-4 DPUs to continuously monitor artificial intelligence agent activity. Sentry can intervene if an agent attempts to move beyond its defined software boundary and can quarantine and stop the agent in milliseconds.
Boitano stated that OpenShell governs the agent actions, while Sentry independently monitors and contains suspicious behavior. More than 100 companies are working with the system at its launch, including Microsoft, Perplexity, Accenture, and JPMorgan Chase.
Because OpenShell is open source, Nvidia said it can also be extended to work with third-party computing platforms, including those from Arm and Intel.
