Not by quarantine alone. NVIDIA says its Open Agent Safety Platform can restrict an AI agent’s activity with OpenShell policies and contain a running agent with Sentry. But stopping an agent is different from invalidating credentials it may have copied or undoing actions it has already completed. Those steps depend on the systems that issued the credentials and handled the agent’s requests.
What NVIDIA’s platform does
NVIDIA announced the Open Agent Safety Platform on September 28, 2026. It pairs OpenShell, open-source runtime software, with Sentry, a reference design for monitoring and containing agents. NVIDIA describes the combination as governance across software, compute, and robotics systems.
As an Amazon Associate I earn from qualifying purchases.
OpenShell operates at the agent runtime boundary. NVIDIA says it runs agents in sandboxes and enforces operator-defined limits on files, networks, tools, processes, and credentials. Those limits are checked before execution and enforced as the agent works. See NVIDIA’s platform announcement and NVIDIA’s technical blog for the vendor’s descriptions.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Sentry is described as an out-of-band watchdog running on NVIDIA BlueField-4 DPUs. NVIDIA says Sentry and DOCA inspect agent requests and responses, provide telemetry, verify identity, and enforce granular access policies for data, tools, APIs, and services. The company says Sentry can quarantine an agent that attempts to leave its boundary, in milliseconds. That timing is NVIDIA’s claim; the sources cited here do not provide an independent benchmark.
#1 Best Overall
What “revoking access” can mean
The phrase can refer to three different controls. A deployment may have one without automatically providing the others:
- Block a future action: Runtime policies can deny operations that pass through the configured enforcement boundary, such as access to a restricted file or network destination.
- Contain the running agent: NVIDIA says Sentry can quarantine an agent after a boundary violation. This is intended to stop or isolate the workload; it does not, by itself, establish that a service elsewhere has invalidated a credential.
- Invalidate credentials or reverse effects: A token must be revoked by its issuer, and active sessions may need to be terminated at the connected service. Completed API calls, messages, file changes, or transactions are separate effects; stopping the agent does not automatically undo them.
NVIDIA’s announcement and technical materials describe policy enforcement, identity verification, and quarantine. They do not establish universal invalidation of credentials already copied into third-party systems, termination of every external session, or rollback of completed actions. Do not treat “quarantined” as equivalent to “all access revoked.”
Rank #2
Does OpenShell require BlueField hardware?
No. NVIDIA product materials describe OpenShell as deployable without BlueField-4. OpenShell provides software-level runtime controls; Sentry is the additional hardware-backed containment layer in NVIDIA’s reference design. A team can therefore consider OpenShell in a software-only environment, while Sentry’s described design requires BlueField-4 hardware.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallWhat a deployment must verify
Whether access can truly be revoked depends on where credentials live and where requests are checked. Before relying on an agent control, determine:
Rank #3
- Whether secrets are held by a gateway or exposed to the agent process, and whether the agent can copy or persist them.
- Whether every relevant request passes through the policy enforcement point, or whether the agent can reach a service by another route.
- Who can revoke each token, terminate sessions, or disable delegated authority at the issuing service.
- What happens to in-flight requests when the agent is blocked or quarantined.
- Whether logs capture policy decisions and enough information to establish whether data left the controlled boundary, while protecting sensitive log contents.
The NVIDIA materials do not specify the answer to these questions for every deployment. They must be resolved for the actual credentials, services, and enforcement paths in use.
Why agent containment is only one security layer
NVIDIA’s NeMo Agent Toolkit security guidance identifies risks including tool misuse, unauthorized data access, unintended API calls, command execution, resource exhaustion, data leakage, and credential exposure. Its recommended practices include role-based access control, rate limiting, sandboxing or containerization, least-privilege access, secret management, log scrubbing, and access controls for logs. These are general safeguards, not proof that a particular OpenShell or Sentry setup is secure by default. See NVIDIA’s NeMo Agent Toolkit security guidance.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How to assess an agent-security control
When comparing runtime controls, gateways, or hardware-backed monitoring, focus on the enforcement path rather than the label:
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsQuick Recap
- Where it enforces: Inside the agent harness, at runtime, at a gateway, or in an independent hardware-backed layer.
- What it covers: Which files, tools, APIs, services, identities, and credentials are actually in scope.
- Whether it blocks or detects: Can the control deny a request before execution, or does it only report suspicious behavior?
- What it can revoke: Can it block future requests only, or can it also invalidate credentials and terminate sessions at their issuers?
- What happens during interruption: How in-flight work is handled and whether already completed effects need separate remediation.
- How it is audited: Whether operators can review enforcement decisions and investigate possible data movement.
- What the evidence supports: Whether performance and containment claims are vendor descriptions or independently tested results.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
