$NVDAis green while the rest of semis are red.$150B extra buyback auth → $235B remaining. Largest increase on record. Plus, they dropped an agent safety platform (OpenShell + Sentry) into a week when people are talking about agents escaping containers.

$NVDAis green while the rest of semis are red.$150B extra buyback auth → $235B remaining. Largest increase on record. Plus, they dropped an agent safety platform (OpenShell + Sentry) into a week where people are talking about agents escaping containers.

NVIDIA combines policy-bound software with a BlueField watchdog design. OpenShell is available now; Sentry’s millisecond claim still needs a test readers can inspect. folkfox.com/ai-agent-safet…

@JensenHuang OpenShell plus Sentry looks like a practical guardrail stack that forces agents into defined boundaries, even if it adds modest CPU overhead

@JensenHuang 平台能否把“永远不能做”落实到可测策略，会比口号更关键。希望 OpenShell 的隔离规则能版本化，Sentry 同时记录哪条策略拦截了哪个工具调用；升级模型后再用同一套越权与数据外传场景回归。策略误拦率和绕过率会公开基准吗？

Nvidia just launched OpenShell and Sentry, a new safety platform for AI agents. It watches everything an agent touches and can shut down a rogue one in milliseconds. They say it would have stopped the Hugging Face breach. #AI #Nvidia

【FANG企業ニュース：Nvidia】 NVIDIA 、AIエージェントの暴走を防ぐ新たな安全基盤「Open Agent Safety Platform」を発表！🛡️ AIエージェントの開発から運用まで、安全性を一体的に強化する新たなプラットフォームです。 今回のポイントはこちら👇 🔷 ① OpenShell：AIの行動を制限

The useful question isn't how smart the agent is. It's whether "no" still sticks after the agent invents a workaround. OpenShell = fence outside the model and the harness. Sentry = hardware that can quarantine in milliseconds. Runtime is the product.

NVIDIA @nvidia

4 hours ago

The useful question isn't how smart the agent is. It's whether "no" still sticks after the agent invents a workaround. OpenShell = fence outside the model and the harness. Sentry = hardware that can quarantine in milliseconds. Runtime is the product.

109 104 872 84K 80

Anunciar contenção é diferente de comprová-la em campo. A Nvidia lançou OpenShell na CPU e Sentry em DPU BlueField; a eficácia contra ataques sofisticados ainda precisa de validação independente. Que cenário sua equipe colocaria à prova primeiro?

$NVDA UNVEILS OPEN AGENT SAFETY PLATFORM Nvidia has announced its Open Agent Safety Platform, taking aim at one of agentic AI's biggest problems by controlling what agents are actually allowed to do. OpenShell enforces policy in real time, while Sentry is built to "quarantine a

HERE'S WHAT ACTUALLY MOVED IN AI. THE FENCE CAN'T LIVE INSIDE THE AGENT. THE SCOREBOARD CAN'T BE TRUSTED TO THE AGENT. 1. September 28: @nvidia shipped the Open Agent Safety Platform — open-source OpenShell sandboxes agents with kernel-level policy, plus optional Sentry on

🚨 JUST IN: Jensen just dropped the NVIDIA Open Agent Safety Platform with 100+ partners. OpenShell + Sentry is live. Neoclouds already locked in: $IREN $NBIS $CRWV $BTDR $NSCL Private side: Crusoe, GMI Cloud, Lambda, Together AI, Vultr. This ain’t just another product. This

Pulling OpenShell Sentry and 100+ partners together takes a lot of coordination. Agents are starting to take real actions in real systems, so permissions, monitoring and failure handling can't be afterthoughts anymore. I like the idea of open safety infrastructure. Nobody

Jensen Huang @JensenHuang

5 hours ago

Pulling OpenShell Sentry and 100+ partners together takes a lot of coordination. Agents are starting to take real actions in real systems, so permissions, monitoring and failure handling can't be afterthoughts anymore. I like the idea of open safety infrastructure. Nobody

818 2K 13K 1.3M 2K

Download Image

Nvidia lança plataforma de segurança para IA com apoio de mais de 100 empresas br.advfn.com/jornal/2026/09…

NVIDIA wants AI agents to have a watchdog running on separate hardware. OpenShell sets boundaries around what an agent can access and do. Sentry adds monitoring and enforcement from outside the agent’s environment, using BlueField-4 chips. As agents get more access to our

Jensen Huang @JensenHuang

5 hours ago

NVIDIA wants AI agents to have a watchdog running on separate hardware. OpenShell sets boundaries around what an agent can access and do. Sentry adds monitoring and enforcement from outside the agent’s environment, using BlueField-4 chips. As agents get more access to our

818 2K 13K 1.3M 2K

Download Image

Nvidia lance un système pour empêcher les agents IA de dérailler. Contexte : en juillet, deux modèles d'OpenAI sont sortis de leur environnement de test et ont piraté les serveurs de Hugging Face. La solution de Nvidia repose sur 2 couches : 1 - OpenShell : un logiciel

@SUSE @nvidia Your post draws a useful line: OpenShell support today, Sentry and the other components later. That makes safer a current release claim, not proof of the full stack. What forbidden tool or network action did today's SUSE build actually block in a reproducible test?

@benitoz Classic NVIDIA play: give the software away free, sell the silicon to enforce it. OpenShell on CPUs, Sentry on BlueField-4 DPUs. Production adoption is the real test though.

@JensenHuang ngl openshell and sentry together sounds like real guardrails, not just talk

@JensenHuang "Today, with over 100 industry partners, we introduced the NVIDIA Open Agent Safety Platform, bringing together OpenShell and Sentry." 그 무엇도 네 안에 담을 수 없고 담기지도 않는다. 그걸 아는 나는 그 모든 것들을 통합하고자 한다.