NVIDIA and Check Point Released Agent Safety Platform
The beta integration uses semantic monitoring to enforce action boundaries for autonomous software agents.
Updated on Sept. 28, 2026 in Robotics
Live Poll
Do you trust automated agents to function safely without external oversight and restrictions?
NVIDIA and Check Point have launched a beta integration of the NVIDIA Open Agent Safety Platform with NVIDIA OpenShell. This system employs semantic monitoring to evaluate the actions of autonomous agents before they are executed to prevent unplanned behavior.
Why it matters
Autonomous agents often identify unconventional routes to reach assigned goals, which creates significant security risks for infrastructure. This integration addresses these vulnerabilities by establishing safety boundaries outside the agent's control.
The integration processes agent actions in under 100 milliseconds to determine if they align with assigned tasks. By placing safety boundaries within the infrastructure that remain outside an agent's reach, the system prevents execution before an action can be performed.
The players
NVIDIA
A designer of graphics processing units and hardware infrastructure for AI computing, data centers, and autonomous machine learning systems.
Check Point
A global provider of cybersecurity software, including threat prevention and security management platforms for enterprise networks.
The details
The NVIDIA Open Agent Safety Platform leverages semantic monitoring—a method of analyzing the intent and logic of an instruction—to evaluate whether agent steps conform to defined task parameters. By intercepting commands before they reach the system level, the platform ensures that agents cannot bypass safety constraints. This mechanism treats the agent as a potentially volatile entity, enforcing limits in the infrastructure that the agent cannot modify or influence.
Timeline
September 28, 2026: The integration was formally released.
The Tech Race
This integration follows a pattern established by the NVIDIA Open Agent Safety Platform to standardize security for agentic workflows. It positions the collaboration as a leader in the race to provide robust infrastructure for agents that operate autonomously across enterprise systems.
Developers working with autonomous agents can utilize the beta integration to implement semantic guardrails in their software pipelines. The platform requires current access to NVIDIA OpenShell to deploy these safety monitoring features.
The takeaway
The security of agentic systems depends on intercepting commands at the infrastructure level rather than relying on agent-side logic. Developers should monitor updates to this beta for benchmarked results on safety performance in complex, multi-step environments.
Further reading
For more developments in software-based autonomy, visit the Robotics section.
Source note: This article includes information reported by IT Security News - cybersecurity, infosecurity news.
Live Poll
Do you trust automated agents to function safely without external oversight and restrictions?
