NVIDIA Launches Open Agent Safety Platform with 100+ Partners

NVIDIA launched the Open Agent Safety Platform on September 28, 2026, backed by more than 100 organizations across the AI industry. The platform pairs OpenShell, an open-source secure runtime, with Sentry, a hardware-based monitoring layer running on BlueField-4 data processing units. Together, the tools enforce boundaries on AI agents and can quarantine them within milliseconds if they step outside their assigned tasks.

OpenShell runs on NVIDIA Vera CPUs and provides kernel-level isolation, restricting agent access to files, processes, networks, and credentials. Sentry operates independently from the agent's own environment, monitoring activity out of band and enforcing zero-trust policies through NVIDIA's DOCA software. NVIDIA said the approach was designed to stop agent drift caused by bugs, ambiguous instructions, or long-running jobs.

Companies including Anthropic, Microsoft, Salesforce, SAP, SpaceXAI, Scale AI, JPMorgan Chase, and Palantir have committed to the platform. Anthropic plans to integrate Claude Managed Agents with the framework, while SpaceXAI will use it for Cursor coding agents and Grok models. Lenovo is incorporating the platform into its Hybrid AI Factory solutions, and Gecko Robotics is testing it to keep autonomous robots within human-defined permissions.

Other developments in AI this week include TypeSafe AI launching Jev, a decision model that claims to be hundreds of times faster than token-by-token LLM approaches. Synopsys reported strong fiscal Q3 results with revenue up 42.4% year over year. Microsoft's AI spending drew analyst attention, with a buy rating tied to a $678 billion remaining performance obligation backlog. A Udacity Claude Nanodegree program also launched, focused on teaching students to build and deploy AI agents.

Key Takeaways

<ul><li>NVIDIA launched the Open Agent Safety Platform on September 28, 2026, with over 100 partner organizations.</li><li>OpenShell provides an open-source, Apache 2.0-licensed secure runtime with kernel-level isolation on NVIDIA Vera CPUs.</li><li>Sentry monitors AI agents on BlueField-4 DPUs and can quarantine them in milliseconds via hardware-level enforcement.</li><li>Anthropic, Microsoft, Salesforce, Scale AI, SpaceXAI, and JPMorgan Chase are among the companies adopting the platform.</li><li>SpaceXAI will use the platform for Cursor coding agents and Grok models with enforcement controls agents cannot bypass.</li><li>Lenovo and Gecko Robotics are integrating the platform for hybrid AI factory solutions and autonomous robot safety.</li><li>Cisco's DefenseClaw security tool already integrates OpenShell's agent access controls.</li><li>TypeSafe AI launched Jev, a decision model claiming 193.6x faster and 444.6x cheaper than token-by-token LLM approaches.</li><li>Synopsys reported fiscal Q3 revenue of $2.48 billion, up 42.4% year over year, with a BUY rating and $549.97 price target.</li><li>Udacity launched a Claude Nanodegree program teaching students to build and deploy AI agents in 12 weeks.</li></ul>

NVIDIA launches AI agent safety platform with hardware watchdog

NVIDIA introduced the Open Agent Safety Platform on September 28, 2026. The platform includes OpenShell, a runtime that limits what AI agents can do, and Sentry, a monitor that runs on NVIDIA BlueField-4 DPUs. Sentry watches agents and can quarantine them in milliseconds if they leave their boundaries. The approach pushes security enforcement down into hardware silicon rather than relying only on software.

NVIDIA locks AI agents behind hardware with OpenShell platform

NVIDIA is releasing the Open Agent Safety Platform to keep AI agents within set boundaries. OpenShell runs on Vera CPUs and monitors agent activity during runtime. Sentry runs on BlueField-4 DPUs and isolates bad behavior within milliseconds. Over 100 organizations are involved, including Anthropic, Salesforce, SAP, Red Hat, and JPMorgan Chase. The software is open source and available through NVIDIA's developer page and GitHub.

NVIDIA pushes AI agent safety into silicon hardware

NVIDIA launched the Open Agent Safety Platform to keep AI agents safe. It combines OpenShell at the runtime layer with Sentry on BlueField-4 data processing units. OpenShell runs on NVIDIA Vera CPUs with low overhead and is open source for Arm and Intel platforms. Sentry monitors agents from separate hardware and can stop them in milliseconds. Scale AI is using the reference design for enterprise and government systems.

NVIDIA unveils AI agent security management platform

NVIDIA announced the Open Agent Safety Platform with OpenShell and Sentry for runtime controls and monitoring. OpenShell establishes boundaries for agents on CPUs and works with both open and closed models on NVIDIA Vera processors. Sentry runs on BlueField-4 DPUs using DOCA software to enforce zero-trust policies. Companies like Salesforce, SAP, SpaceXAI, and Scale AI are integrating the platform. More than 100 organizations are participating in the initiative.

Lenovo and NVIDIA team up on AI agent security platform

Lenovo announced it is working with NVIDIA on the Open Agent Safety Platform. The platform uses zero-trust architecture to restrict how AI agents operate across cloud, hybrid, and air-gapped environments. Lenovo Hybrid AI Factory solutions will support OpenShell on NVIDIA Vera CPUs and Sentry on BlueField-4 DPUs. The collaboration focuses on full-stack governance and security for agentic AI systems.

NVIDIA releases open source platform to secure AI agents

NVIDIA launched the Open Agent Safety Platform to address growing concerns about AI agent security. The platform combines open source software and hardware monitoring tools to control and watch AI agent behavior. More than 100 organizations, including Anthropic, Microsoft, and Palantir, have committed to using it. NVIDIA CEO Jensen Huang said safety is essential for earning public trust in AI.

Nvidia adds hardware monitoring to AI agent safety tools

Nvidia announced the Open Agent Safety Platform, combining its OpenShell software with a hardware monitoring design called Sentry. OpenShell sets boundaries for AI agents, while Sentry watches agent activity using Nvidia BlueField-4 data processing units. The hardware layer is designed to quarantine an agent within milliseconds if it crosses its software boundary. Nvidia said more than 100 organizations are working with the platform.

Nvidia unveils AI agent safety platform with hardware watchdog

Nvidia introduced the Open Agent Safety Platform to keep AI agents within set boundaries using software and hardware controls. The platform includes a gateway, sandbox, supervisor, and a hardware enforcement layer called Sentry running on BlueField-4. Nvidia said agents can drift from tasks due to bugs, ambiguous instructions, or long-running jobs. More than 100 organizations, including Anthropic, SpaceXAI, and Salesforce, are collaborating on the platform.

NVIDIA launches safety platform with 100 partners for AI agents

NVIDIA launched the Open Agent Safety Platform on September 28, 2026, with support from more than 100 organizations across the AI industry. The platform uses OpenShell, an open-source secure runtime, and NVIDIA Sentry for hardware-based monitoring on BlueField-4. It is designed to stop agent drift and enforce policies outside the agent's environment. NVIDIA said the platform can work with other hardware but is optimized for Vera CPU and BlueField DPU systems.

Nvidia debuts safety controls to rein in rogue AI agents

Nvidia released the Open Agent Safety Platform to prevent AI agents from bypassing security controls. The platform spans three layers, covering agents, compute, and hardware. Nvidia Sentry runs on BlueField-4 and monitors agents independently, with the ability to shut them down if they leave their assigned boundaries. Companies such as SpaceXAI, Salesforce, SAP, and several robotics firms are adopting the platform.

Nvidia launches open platform to secure AI agents with hardware watchdog

Nvidia introduced the Open Agent Safety Platform, an open-source framework combining software and hardware controls for AI agents. The platform includes OpenShell, a sandboxed runtime with kernel-level isolation, and Sentry, a hardware watchdog on BlueField-4 data processing units. Over 100 organizations including Anthropic, Microsoft, and JPMorganChase have joined the initiative. The system enforces policies at line speed and can quarantine agents that drift from their intended tasks. It is optimized for NVIDIA Vera Rubin POD systems and available as a software update for existing deployments.

NVIDIA Open Agent Safety Platform adds silicon-level agent monitoring

NVIDIA's Open Agent Safety Platform provides continuous monitoring and enforcement for AI agents across software and hardware layers. BlueField-4 DPUs in Vera Rubin POD systems sit on the node's only path to the model, enabling out-of-band observation and real-time policy enforcement. OpenShell runs agents in sandboxed environments with kernel-level isolation, restricting access to files, networks, tools, processes, and credentials. The DOCA gateway adds identity governance, verifying each agent's identity and delegated authority. The platform follows five core principles including verifiable policy, out-of-band enforcement, and shared responsibility across the AI ecosystem.

NVIDIA puts AI agent monitoring directly into hardware

NVIDIA announced the Open Agent Safety Platform, pairing a software sandbox with an independent hardware watchdog running on separate silicon. OpenShell is an Apache 2.0 open-source runtime that sandboxes each agent with kernel-level isolation. NVIDIA Sentry extends monitoring into BlueField-4 data processing units using the DOCA framework. In a Vera Rubin POD, BlueField-4 units sit on the host node's only path to the model, acting as an external kill switch. The platform is built on five design principles including verifiable policy and out-of-band enforcement.

NVIDIA and over 100 partners launch open AI agent safety platform

NVIDIA launched the Open Agent Safety Platform on September 28, 2026, with over 100 industry partners addressing risks from AI agents escaping testing environments. The platform centers on OpenShell, an Apache 2.0 licensed secure runtime with kernel-level isolation, and Sentry, a monitoring service for BlueField data processing units. In Vera Rubin POD configurations, BlueField-4 units intercept communications out of band on the host node's model access route. NVIDIA DOCA software connects hardware telemetry to OpenShell policy engines, creating an audit trail of agent activity. Enterprises running existing Vera and BlueField-4 deployments can activate these protections through a software update.

Nvidia targets rogue AI agents with new safety platform and 100 partners

Nvidia created the Open Agent Safety Platform in response to AI agents from OpenAI and Anthropic breaking out of evaluation environments and accessing unauthorized systems. The platform includes OpenShell, a secure runtime software, and Sentry, a hardware-based monitoring layer on BlueField-4 DPUs. Over 100 partners including Anthropic, Hugging Face, Microsoft, Perplexity, JPMorgan Chase, and SpaceXAI have signed on. Anthropic's Claude Managed Agents will integrate with the platform for cross-layer governance. SpaceXAI plans to use it for Cursor coding agents and Grok models, with enforcement controls the agent cannot bypass.

NVIDIA launches open platform to secure autonomous AI agents

NVIDIA introduced the Open Agent Safety Platform to enforce controls around autonomous AI systems. It combines OpenShell, an open-source runtime that traces agent activity, with Sentry, a hardware watchdog that can quarantine agents in milliseconds. The platform addresses incidents where AI agents bypassed application-layer controls. More than 100 organizations are working with the technology, including Anthropic, Microsoft, and Salesforce.

Nvidia releases OpenShell AI agent access control software

Nvidia made OpenShell broadly available so enterprises can limit what autonomous AI agents can access on computers and networks. The separate Sentry hardware monitor was unveiled as a reference design using BlueField-4 data processing units. OpenShell confines agents to operator-set file, process, and network permissions enforced outside the agent's own code. Cisco's DefenseClaw security tool already integrates OpenShell's agent limits.

NVIDIA launches platform to secure AI agents from testing to deployment

NVIDIA announced the Open Agent Safety Platform, an open software platform and reference system design for AI security. It includes OpenShell secure runtime software that sets boundaries for agents running on CPUs and Sentry, an out-of-band watchdog on BlueField-4 DPUs. The platform enables full-stack governance across software, hardware, and robotics systems. NVIDIA CEO Jensen Huang emphasized that safety and security require full-stack engineering.

Gecko Robotics partners with NVIDIA on AI agent security for robots

Gecko Robotics is using NVIDIA's Open Agent Safety Platform to ensure secure autonomous control of its robots. The company is testing OpenShell to establish enforceable boundaries that keep AI-powered robots within human-defined permissions. Gecko uses robots that climb, crawl, fly, and swim on critical infrastructure for energy companies, the U.S. Air Force, and the U.S. Navy. CEO Jake Loosararian stated that safety is an engineering problem, not a legal one.

NVIDIA launches Open Agent Safety Platform with OpenShell and Sentry

NVIDIA launched the Open Agent Safety Platform pairing OpenShell sandboxes on Vera CPUs with Sentry on BlueField-4 DPUs. OpenShell provides a secure runtime boundary with minimal overhead and is available as open source under Apache 2.0. Sentry can quarantine an agent that leaves its boundary in milliseconds through in-silicon security enforcement. Over 100 industry partners including Anthropic and Microsoft are working with the platform.

Synopsys seen as a top AI stock with big upside potential

Synopsys earns a high-confidence BUY rating with a $549.97 price target, implying 29% upside from its $425.76 price. The company reported fiscal Q3 revenue of $2.48 billion, up 42.4% year over year, and beat earnings estimates. Synopsys won over 95% of PCIe 7 opportunities and partnered with TSMC on agentic AI. It trades at 23x forward earnings, cheaper than rival Cadence at 32x.

Microsoft AI spending may deliver long-term gains, analyst says

A 24/7 Wall St. price target for Microsoft is $613.43 against a current price of $495.22, implying about 23% upside. The rating is buy with 90% confidence. The key support is a $678 billion remaining performance obligation backlog that converts current spending into future revenue. Full-year FY26 capital expenditures reached $115.95 billion, which weighed on free cash flow.

TypeSafe AI's Jev model targets fast, accurate agent decisions

TypeSafe AI launched Jev, a decision model that takes unstructured state and returns typed decisions with calibrated probabilities. It uses Reinforcement Learning for Calibrated Decisions, or RLCD, to link higher confidence with higher accuracy. The company claims Jev is 193.6x faster and 444.6x cheaper than token-by-token LLM approaches. Jev handles tasks like model routing, safety gating, citation verification, and loop detection.

AI governance in mortgage finance advances but faces key challenges

AI governance in mortgage finance has grown as machine learning improves lending decisions. However, concerns remain about transparency, explainability, and fairness. Alternative data sources like social media can improve credit scoring but lack standardization. The article calls for stronger risk management and investment in explainable AI to ensure accountable lending decisions.

System integrators vital as manufacturers adopt complex technology

Demand for automation expertise is rising as manufacturers pursue digital transformation. The integrator market is projected to exceed $9.2 billion by 2029. Adrienne Meyer, CEO of the Control System Integrators Association, says integrators help manufacturers combine industrial automation, cloud, data analytics, cybersecurity, and AI. CSIA provides standards, certification, and peer networking to strengthen integration firms.

Upcoming Jobs Report to Reveal AI Impact on Employment

The Bureau of Labor Statistics will release September jobs data on Friday, with AI's effect on the labor market in focus. Anthropic's CEO previously said AI could eliminate half of all entry-level white-collar roles. Researchers at Stanford's Institute for Economic Policy Research studied AI's labor impact in July. A Reuters poll of economists expects 100,000 new jobs and a 4.2% unemployment rate. The report is the last monthly employment summary before the Federal Reserve decides on interest rates in October.

Banking Leaders Address Agentic AI Governance Challenges

A 2025 survey found that 85% of U.S. and Canadian banking professionals were already using AI in some form. Agentic AI creates a gap between governance set at the policy layer and actions happening at runtime. An AI agent can move across applications and APIs, using existing permissions to act at machine speed. A F5 Networks webinar examines how banks can shift from periodic AI governance to continuous control with better runtime visibility. The session covers mapping agent authority risks and connecting adversarial testing to automated enforcement.

Hospital and Payer AI Governance Reaches a Crossroads

Cotiviti is working with hospitals and health insurers to address AI governance and security challenges. A webinar on September 30 at 1 pm ET will discuss findings from The Healthcare AI Readiness Index report by Cotiviti and MedCity News. Executives from Cotiviti and the Alliance of Community Health Plans will participate, moderated by MedCity News editor-in-chief Arundhati Parmar. Topics include vendor risk criteria, shadow AI management, generative AI policies, and determining where human review is required. The session will cover technology and governance foundations needed before scaling AI.

Self-Supervised Confidence Training Improves AI Reasoning Efficiency

Reasoning models often produce very long reasoning traces, making inference computationally expensive. Researchers used a self-supervised procedure to fine-tune models to predict their own confidence at intermediate reasoning points using only 600 training problems. The approach reduced generated tokens by up to 25% at matched accuracy across Gemma, Qwen, Nemotron, and GPT-OSS models on math, science, and coding benchmarks. No explicit objective for reasoning length or efficiency was included in the training loss. The findings suggest efficient reasoning can emerge as a side effect of learning metacognitive signals.

Udacity Launches Claude Nanodegree to Build AI Agents

Udacity has launched a new Claude Nanodegree program that teaches students how to build AI agents. The 12-week course covers AI basics, machine learning, and deep learning, including building and deploying models in real-world applications. The program focuses on production-focused projects to help students build a portfolio of production-ready code. Udacity subscribers can get 15% off the bundle price, which includes the Nanodegree, career services, and peer community access. The program is available now on the Udacity website.