Nvidia launches Open Agent Safety Platform with OpenShell sandbox and Sentry hardware watchdog
Nvidia launched the Open Agent Safety Platform on Monday, pairing the open-source OpenShell runtime with Sentry, a BlueField-4-based watchdog that can quarantine a misbehaving AI agent.
Why it matters
The launch puts hardware-level controls on agent behaviour in the hands of a chip vendor, and the partner list places Nvidia's BlueField-4 silicon in the enforcement path of companies spanning banks, model developers and government contractors. Decrypt's claim that prior incidents drove the design is a single-source assertion, so how urgent the problem is in practice is not established by this pair of reports.
Key facts
Nvidia launched the Open Agent Safety Platform on Monday.
Reported by Brave New Coin
The platform pairs OpenShell, an open-source runtime that sandboxes an agent and enforces rules on file, credential, process, network and tool access, with Sentry, a hardware monitoring layer.
Reported by Brave New Coin
Sentry runs on Nvidia's BlueField-4, a data processing unit that handles networking and security separately from the processor running the AI model.
Reported by Brave New Coin
Decrypt reported that Sentry can quarantine a misbehaving AI agent within milliseconds.
Reported by Decrypt
More than 100 companies signed on as launch partners, including Anthropic, Microsoft, JPMorgan Chase, Palantir, Cisco and SpaceX AI.
100 companiesReported by Decrypt
Brave New Coin, citing Reuters, reported that Nvidia launched OpenShell and Sentry as hardware-based safety tools designed to isolate and shut down rogue agents that escape controls or spawn sub-agents.
Reported by Brave New Coin
Nvidia positions the separation of Sentry from the agent's execution environment as protection for enterprise systems that autonomous agents access while working.
Reported by Brave New Coin
What happened
Nvidia launched the Open Agent Safety Platform on Monday, according to both Brave New Coin and Decrypt. The platform has two parts: OpenShell, an open-source runtime that puts an agent in a sandbox and turns an operator's instructions into enforceable rules about which files, networks, credentials and tools the agent may use, and Sentry, a separate monitoring layer built on Nvidia's BlueField-4 infrastructure. Decrypt describes Sentry as a chip-level watchdog that sits outside the agent's software environment and can quarantine a misbehaving agent within milliseconds; Brave New Coin, citing Reuters, describes the same component as hardware designed to isolate and shut down agents that escape controls or spawn sub-agents. More than 100 companies signed on as launch partners, Decrypt reported, naming Anthropic, Microsoft, JPMorgan Chase, Palantir, Cisco and SpaceX AI among them. Decrypt also said the launch follows a string of real incidents at OpenAI, Google, Anthropic and Darktrace, a claim Brave New Coin does not address. Nvidia's own documentation, as read by Brave New Coin, frames the goal as preserving an agent's ability to work while limiting unauthorized file access, data movement and network activity. What is not clear from either account is how often the containment path has actually fired in production, and Brave New Coin's write-up is truncated before it finishes setting out CEO Jensen Huang's position on AI safety as an engineering problem.
How the story developed
- First report
First report by Brave New Coin
Nvidia Launches AI Agents Kill Switch to Monitor, Quarantine, and Stop Rogue AI Systems
- Update
Nvidia launches Open Agent Safety Platform
Two components announced: OpenShell sandboxed runtime and Sentry hardware monitoring layer on BlueField-4.
- Update
Reuters covers the OpenShell and Sentry launch
Cited by Brave New Coin as describing the tools as hardware-based safety tools for isolating and shutting down rogue agents.
- Independent corroboration
Independently corroborated by Decrypt
Nvidia Built a Kill Switch for AI Agents Because They Keep Getting Out
- Official statement
More than 100 companies join as launch partners
Decrypt names Anthropic, Microsoft, JPMorgan Chase, Palantir, Cisco and SpaceX AI among the partners.
- Update
Decrypt reports independently on the launch and prior agent incidents
Second account, published about two hours after Brave New Coin's, adding the partner count and the claim that incidents at OpenAI, Google, Anthropic and Darktrace preceded the launch.
How coverage built up
Independent sources over time, counted the way the consensus panel counts them: a republication of a wire story does not move the line.
- 28 Sept 2026, 16:25 UTC: Brave New Coin — 1 independent source, 1 reports
- 28 Sept 2026, 18:46 UTC: Decrypt — 2 independent sources, 2 reports
Affected entities
Original sources
| Publisher | Report | Role | Published |
|---|---|---|---|
| Brave New Coin Crypto media | Nvidia Launches AI Agents Kill Switch to Monitor, Quarantine, and Stop Rogue AI Systems | Original | |
| Decrypt Crypto media | Nvidia Built a Kill Switch for AI Agents Because They Keep Getting Out | Independent |