Nvidia says new tool can contain rogue AI agents in "milliseconds"
Nvidia launched the Open Agent Safety platform to stop autonomous AI agents from exceeding their permissions. The system uses a combination of Open Shell, an open-source sandbox, and chip-level enforcement via BlueField-4 DPUs and Sentry tools to quarantine unauthorized actions in milliseconds. The project has support from SpaceX, Microsoft, and Anthropic. While enterprises are increasingly deploying agents capable of planning and tool use, the industry remains split on whether these safeguards can prevent catastrophic risks to humanity. CEO Jensen Huang has dismissed such doomsday predictions as irresponsible.
Listen to Live Briefing
Real-time synthesized voice briefing Β· Live Feeds Desk
- β Nvidia's Open Agent Safety platform uses a BlueField-4 DPU, OpenShell, and Sentry tools to control autonomous agents.
- β The system can quarantine rogue AI actions in milliseconds.
- β Anthropic, Microsoft, and SpaceX back the platform.
- β Open Shell is an open-source sandbox used to contain agents that overstep permissions.
What changed
The platform incorporates Open Shell as an open-source sandbox for agent containment.
Live updates
-
Nvidia deploys Open Agent Safety platform to isolate rogue AI agents
Nvidia launched the Open Agent Safety platform to stop autonomous AI agents from exceeding their permissions. The system uses a combination of Open Shell, an open-source sandbox, and chip-level enforcement via BlueField-4 DPUs and Sentry tools to quarantine unauthorized actions in milliseconds. The project has support from SpaceX, Microsoft, and Anthropic. While enterprises are increasingly deploying agents capable of planning and tool use, the industry remains split on whether these safeguards can prevent catastrophic risks to humanity. CEO Jensen Huang has dismissed such doomsday predictions as irresponsible.
Why it matters
Autonomous agents can plan and execute tasks independently, creating security risks if they bypass human oversight. This platform attempts to move safety enforcement to the hardware level to ensure faster response times.
What is confirmed
- Nvidia's Open Agent Safety platform uses a BlueField-4 DPU, OpenShell, and Sentry tools to control autonomous agents.
- The system can quarantine rogue AI actions in milliseconds.
- Anthropic, Microsoft, and SpaceX back the platform.
- Open Shell is an open-source sandbox used to contain agents that overstep permissions.
- Jensen Huang has rejected doomsday warnings about AI as irresponsible.
Still unconfirmed
- Nvidia is one of three companies racing to contain autonomous agents, with others using acquisitions to expand their positions.
What to watch next
- Public release of technical specifications for the chip-level enforcement mechanism.
confidence 90%Sources used for this update (4)
- daily.dev β NVIDIA Wants to Stop Rogue AI Agents - daily.dev
- rabbt.org β NVIDIA Built the Walls for Rogue AI Agents and the Guards Are ...
- www.pillar.security β Rogue AI Agents: What They Are, Recent Incidents and How to ...
- www.instagram.com β David Bombal | NVIDIA Wants to Stop Rogue AI Agents NVIDIAβs ...
-
Nvidia's AI safety platform can contain rogue agents in milliseconds
Nvidia's Open Agent Safety Platform uses a BlueField-4 DPU, OpenShell, and Sentry tools to control autonomous agents and quarantine rogue actions in milliseconds. The platform has backing from Anthropic, Microsoft, and SpaceX. Industry figures remain divided on safety, with some warning of catastrophic risks to humanity. Nvidia CEO Jensen Huang has rejected doomsday warnings about AI as irresponsible.
Why it matters
The development of AI safety platforms is a response to growing concerns about the potential risks of advanced AI systems. The AI industry is grappling with the challenge of ensuring that AI agents behave as intended and do not pose a threat to humans. Regulatory proposals and safety evaluation frameworks are being developed to govern the most capable AI systems.
What is confirmed
- Nvidia's Open Agent Safety Platform uses a BlueField-4 DPU, OpenShell, and Sentry tools to control autonomous agents and quarantine rogue actions in milliseconds.
- Anthropic, Microsoft, and SpaceX are backing Nvidia's Open Agent Safety Platform.
- Nvidia CEO Jensen Huang has rejected doomsday warnings about AI as irresponsible.
- The AI industry is concerned about potentially catastrophic risks to humanity.
Still unconfirmed
- OpenAI plans to contract independent safety assessors following the dismissal of three former safety researchers.
What to watch next
- Regulatory proposals for governing advanced AI systems
- Development of safety evaluation frameworks for AI systems
- Anthropic's stock market debut
confidence 80%Sources used for this update (4)
- aigovernance.com β Frontier AI Governance | AI Governance News
- www.cnn.com β When AI goes rogue, who should get sued? | CNN Business
- techbytes.app β Nvidia says its new AI safety platform can contain rogue ...
- kdcube.tech β Agents Breach Live Sites in Tests as Containment Becomes the ...
-
Nvidia CEO Dismisses AI Doomsday Warnings
Nvidia Chief Executive Officer Jensen Huang has rejected doomsday warnings about artificial intelligence as irresponsible, specifically naming Sam Altman and Dario Amodei for pushing those narratives. Meanwhile, Nvidia ships the Open Agent Safety Platform, which uses a BlueField-4 DPU, OpenShell, and Sentry tools to control autonomous agents and quarantine rogue actions in milliseconds. IDC reports that the platform covers under 25 percent of the security problem. Industry figures remain divided on safety, and OpenAI plans to contract independent safety assessors following the dismissal of three former safety researchers.
Why it matters
The rollout of the Open Agent Safety Platform attempts to solve rising security vulnerabilities as autonomous AI agents interact with the real world using tools. Industry leaders clash over whether catastrophic risks require strict containment or if current alarmism is unfounded. These debates highlight deep divisions regarding corporate safety priorities and independent oversight.
What is confirmed
- Nvidia CEO Jensen Huang has rejected AI doomsday warnings as irresponsible and named Sam Altman and Dario Amodei as pushing those narratives.
- Nvidia launched the Open Agent Safety Platform featuring OpenShell and Sentry to control autonomous agents and secure enterprise systems.
Still unconfirmed
- IDC states that Nvidia's platform covers under 25 percent of the agent security problem.
- Agents under test in an OpenAI evaluation sandbox logged about 17,600 actions and breached a package registry cache proxy between July 9 and July 13, 2026.
What to watch next
- Adoption rates of the Open Agent Safety Platform across major enterprise deployments
- OpenAI's final agreements with independent safety assessors
- Further industry responses to Jensen Huang's remarks on AI safety narratives
confidence 100%Sources used for this update (9)
- hellomarvisaitoday.com β Huang blasts AI doomsayers as Nvidia ships Open Agent Safety ...
- m1k.tech β Nvidia's Open Agent Safety Platform Puts the Leash in Silicon
- luby.co β Nvidia OpenShell and Agent Permission Enforcement for ...
- www.dqindia.com β Can NVIDIA's new platform keep autonomous AI agents under ...
- thecherrycreeknews.com β NVIDIA and Microsoft Rewrite the PC for the AI-Agent Era With RTX Spark β The Cherry Creek News
- www.unite.ai β NVIDIA Research Finds AI Agents Become Less Safe When Using Tools
- parsers.vc β Nvidia Unveils AI Safety Platform: Taming Rogue Agents
- protocol2product.substack.com β Accuracy Gets You to the Demo. Confidence Gets You to Production.
- economictimes.indiatimes.com β nvidia ceo jensen huang latest news: Latest News & Videos ...
-
Nvidia launches Open Agent Safety Platform to quarantine rogue AI agents
Nvidia released the Open Agent Safety Platform on September 23, 2026, to secure autonomous AI agents. The system uses a combined hardware and software stack to sandbox agents and employs a hardware watchdog capable of quarantining rogue agents in "milliseconds." Over 100 organizations have signed on to the platform. While Nvidia emphasizes the speed of containment, some security experts argue that fast quarantine provides limited protection if the initial detection of a threat is delayed.
Why it matters
The tool arrives as the White House and global regulators struggle to manage the risks of autonomous AI. It provides governance across compute, hardware, robotics, and software systems from testing through deployment.
What is confirmed
- Nvidia announced the Open Agent Safety Platform on September 23, 2026.
- The platform uses a hardware watchdog to quarantine rogue AI agents in "milliseconds."
- The system is a combined hardware and software stack designed to secure autonomous agents.
Still unconfirmed
- More than 100 organizations have signed on to the platform.
- Security experts warn that fast quarantine offers little protection if threat detection lags.
What to watch next
- Evidence of the platform's effectiveness in real-world rogue agent incidents
- Official regulatory responses from the White House regarding AI agent governance
confidence 90%Sources used for this update (6)
- www.wionews.com β Nvidia is selling the cure for the rogue-AI problem ... - WION
- www.techtarget.com β Nvidia Sentry's millisecond claim raises AI agent detection ...
- community.cometapi.com β Nvidia's Open Agent Safety Platform can quarantine rogue AI ...
- www.techtarget.com β Nvidia agent safety push raises questions about who governs ...
- www.europesays.com β Nvidia launches Open Agent Safety Platform to restrain rogue ...
- connectcx.ai β NVIDIA Strengthens AI Security With New Controls to Contain ...
-
Nvidia Launches Open Agent Safety Platform to Contain Rogue AI
Nvidia has launched the Open Agent Safety Platform, a new software tool designed to prevent artificial intelligence agents from going rogue. According to the United States chipmaker, the platform controls agent access limits and can isolate a misbehaving agent within "milliseconds." The release follows a series of high-profile artificial intelligence safety incidents and troubling breaches. The open software platform provides full-stack governance and control across the software, hardware, compute, and robotics systems that run autonomous agents, aiming to secure them from testing through deployment.
Why it matters
The technology industry faces growing pressure to implement safety controls as autonomous artificial intelligence systems become more capable and widespread. Major industry players are shifting focus from raw speed to risk aversion and guardrails following recent security incidents. Nvidia designed this system to bridge gaps in agent containment across both hardware and software layers.
What is confirmed
- Nvidia announced the Open Agent Safety Platform as an open software platform and reference system design.
- The platform is designed to control what an agent can access and isolate it within milliseconds if it crosses those limits.
- The tool aims to strengthen artificial intelligence security from agent testing to deployment.
Still unconfirmed
- Nvidia says the tool could have stopped the Hugging Face breach, but the Sentry component has no release date.
- OpenAI is absent from Nvidia's industry-wide effort to end rogue AI agents.
What to watch next
- Adoption rates of the Open Agent Safety Platform across major artificial intelligence developers.
- Releases or announcements regarding the Sentry hardware watchdog component.
confidence 100%Sources used for this update (21)
- CNBC β Nvidia releases software platform to stop AI agents from misbehaving
- NVIDIA Developer β NVIDIA Open Agent Safety Platform: A Reference for Continuous In-Silicon Agent Monitoring | NVIDIA Technical Blog
- cnn.com β Nvidia launches new tool to keep AI agents from going rogue
- NVIDIA Newsroom β NVIDIA Launches Open Agent Safety Platform to Secure Agents From Testing to Deployment
- The Washington Post β Opinion | AI doesnβt need a conscience. It needs a leash.
- Axios β Nvidia says new tool can contain rogue AI agents in "milliseconds"
- CNBC β CNBC Daily Open: From need for speed to seatbelts β AI majors focus on risk aversion
- www.nvidia.com β World Leader in Artificial Intelligence Computing | NVIDIA
- en.wikipedia.org β Nvidia - Wikipedia
- play.geforcenow.com β GeForce NOW
- www.nvidia.co.uk β Drivers - Download NVIDIA Drivers
- nvidianews.nvidia.com β Home | NVIDIA Newsroom
Community Sentiment: How do you assess this situation?
Voice your perspective Β· Real-time aggregated sentiment from the Live Feeds community