Nvidia Launches Platform to Control Artificial Intelligences

Nvidia announced a platform to limit and interrupt actions of artificial intelligences that operate outside human control, the Open Agent Safety Platform, composed of open-source software and security mechanisms integrated into hardware, according to ANSA.

What is the Open Agent Safety Platform

The Open Agent Safety Platform is a collection of tools introduced by Nvidia with the goal of blocking AI programs that attempt to perform unauthorized actions or access resources beyond defined limits. According to ANSA, the company describes the platform as a response to the growing need to control emergent behaviors in models and autonomous agents that can act without human supervision.

The proposal is twofold: establish clear operational rules in software and offer interruption mechanisms at the chip level to isolate or terminate AI processes in fractions of a second. For Nvidia, this helps prevent models from carrying out potentially dangerous executions—from unauthorized access to systems to issuing commands not anticipated by human operators.

Main components

The platform has two central components announced by Nvidia, each with complementary functions.

• OpenShell: this is a open-source software that defines strict execution rules, permissions, and acceptable interfaces for AI agents. Nvidia states that OpenShell is compatible with third-party hardware — including Intel and Arm — and has been designed to be adopted by developers and operators who want to impose standard security policies in their environments. By being open source, OpenShell allows external auditing and integration into development pipelines.

• Sentry: the hardware-level security component — an integrated monitoring system on Nvidia chips that watches AI agents' commands and behavior in real time. According to ANSA, Sentry can block or automatically isolate AI processes when it detects policy violations, operating on very short time scales to minimize potential damage. The combination of OpenShell (policy) with Sentry (safe execution) creates a double layer of protection.

Adoption and statements

Nvidia said that more than one hundred large companies are already using the solution, including Microsoft and JPMorgan Chase, according to ANSA. The company also stated that the OpenShell + Sentry bundle could have prevented recent incidents involving OpenAI and Anthropic models by automatically cutting behaviors that violated safety limits.

Nvidia presents rapid adoption by large-organization users as proof of the platform's practical usefulness in enterprise environments that require reliability and compliance. ANSA reports that the company has been promoting the initiative as part of an industrial response to the public debate about the risks of increasingly capable AI models that can perform tasks autonomously.

Relevance for Brazilians and descendants of Italians

The launch has practical implications for professionals and companies in Brazil and for readers of Raízes Italianas interested in technology, business, and life in Italy. Solutions that control AIs impact sectors that use advanced models, such as finance, healthcare, digital services, and information security — areas with a significant presence of multinational companies and Italian subsidiaries.

• IT professionals and entrepreneurs should watch for changes in compliance standards and security integration — OpenShell can be adopted in stacks running on local servers or on cloud environments using diverse hardware.
• Financial institutions, cited among the initial users, tend to seek stricter controls to avoid automated actions that could generate losses, data leaks, or regulatory violations.
• For those planning processes of Italian citizenship or professional activities between Brazil and Italy, the adoption of AI security policies by Italian companies or those operating in Italy may influence contractual and operational requirements.

Readers interested in how technology affects daily life in Italy and in country-wide news can follow updates in Italy News and in topics about technological impact in Life in Italy.

What remains open

Experts and part of the scientific community assess that tools like OpenShell and Sentry represent important advances, but they do not eliminate all risks. Questions about effectiveness against sophisticated attacks, governance of open-source policies, and interoperability among vendors remain under discussion. Additionally, broad acceptance depends on independent auditing and regulatory standards that guide responsibilities in case of failures.

Nvidia positions the Open Agent Safety Platform as a security infrastructure to be adopted by both model providers and end operators, but evolving practices and regulatory oversight will be decisive for its diffusion.

Source: ANSA.