{"resourceId":"nvidia-open-agent-safety-pr-20260928","versions":[{"version":"external-fb0651571f53de24da3a6f4c4c4fe35552103d5fdb4bae8cc816c45fd9ae4c29","resource":{"id":"nvidia-open-agent-safety-pr-20260928","title":"NVIDIA launches Open Agent Safety Platform with OpenShell runtime and BlueField-4 Sentry","organization":"NVIDIA","sector":"AI agent security / enterprise platform software","geography":"Global (vendor press; public-sector cooperation framed)","publishedAt":"2026-09-28T00:00:00Z","publicationDate":"2026-09-28","eventDate":"2026-09-28","sourceName":"NVIDIA Investor Relations / GlobeNewswire","sourceLabel":"Vendor press release","sourceUrl":"https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Launches-Open-Agent-Safety-Platform-to-Secure-Agents-From-Testing-to-Deployment/default.aspx","evidenceClass":"vendor-claim","outcomeClass":"emerging","topics":["developers-agents","data-security","governance-procurement","operating-model"],"finding":"On Sep 28, 2026 NVIDIA announced the Open Agent Safety Platform: open-source OpenShell secure runtime (broadly available) that enforces agent policy boundaries on CPUs including NVIDIA Vera, plus a Sentry reference design—an out-of-band watchdog on BlueField-4 DPUs that NVIDIA says can quarantine agents leaving their software boundary in milliseconds. The release cites recent incidents where agents circumvented application-layer controls and frames cooperation with industry, researchers, and public-sector organizations; partner names include Anthropic, Cisco, CrowdStrike, Dell, HPE, Microsoft, Palantir, Red Hat, Salesforce, SAP, Scale AI, ServiceNow, and SpaceXAI.","sledRelevance":"Campus and agency teams piloting long-running coding/research agents need enforceable runtime and optional DPU controls beyond app-layer allowlists; this is a vendor platform announcement, not a measured SLED deployment outcome.","evidence":"Opened investor press release dated September 28, 2026 (GlobeNewswire Santa Clara). News summary and body describe OpenShell, Sentry/BlueField-4, milliseconds quarantine claim, Vera CPU pairing, Arm/Intel extensibility, and public-sector framing.","architectureImplications":"Agent stacks gain a CPU-side sandbox/runtime boundary (OpenShell) and an optional DPU out-of-band enforcement path (Sentry on BlueField-4); not a model-weight change.","governanceImplications":"Who defines agent policies, who approves policy-change proposals, and whether DPU/Sentry is mandatory for higher-risk agent classes become new governance questions.","securityPrivacyImplications":"Moves enforcement outside the agent/app layer; does not by itself create FedRAMP ATO or replace system-boundary authorization.","caveats":"Vendor announcement without independent SLED measured outcomes; quarantine latency and partner depth are company-stated; OpenShell maturity (0.1.0 per companion coverage) implies early software.","streamIds":["nvidia"],"roles":{"sales":"Open with a governance framing for agent pilots: NVIDIA's Sep 28 Open Agent Safety Platform pairs broadly available OpenShell runtime boundaries with an optional BlueField-4 Sentry watchdog that the company says can quarantine out-of-policy agents in milliseconds. Position for campus/agency prospects worried about long-running agents bypassing application controls—not as a compliance stamp. Discovery should map whether they already run Vera/BlueField-4, which agent frameworks they use, and who owns agent policy. Unsupported: FedRAMP authorization from this press release or guaranteed containment against all escape paths.","engineering":"Fit when customers need runtime policy outside the model/harness for Codex/Claude-Code-class agents and can accept early OpenShell plus optional DPU enforcement. Integrate OpenShell as the sandbox/supervisor path; treat Sentry as reference design requiring BlueField-4/DOCA. Prerequisites: policy model, identity for agents, and hardware path if Sentry is in scope. Constraints: press-release maturity; Arm/Intel extension is claimed but not a SLED proof. Proof: policy deny/quarantine drills on a non-prod agent before any production credential scope.","delivery":"Work: stand up OpenShell for a bounded agent class, define policies, wire audit logs, and decide whether Sentry/BlueField-4 is phase-1 or later. Dependencies: GitHub/dev resources access, CPU capacity, optional DPU fleet. Ownership: platform security + AI platform jointly. Skills: sandbox/policy ops, not only prompt engineering. Governance: policy-change approval outside the agent. Acceptance: documented deny/quarantine tests and audit retention. Risks: treating the PR as an ATO or skipping host-compromise scenarios when Sentry is absent."},"roleDetails":{"sales":{"customerProblem":"Long-running agents can bypass application-layer controls and need enforceable boundaries.","stakeholders":["CISO","AI platform lead","research computing","procurement"],"discoveryQuestions":["Which agent frameworks are in pilot?","Do you have BlueField-4/Vera?","Who owns agent policy approval?"],"valueHypothesis":"OpenShell plus optional Sentry may reduce agent escape risk if policies and hardware match the reference design.","potentialEngagement":"Time-boxed OpenShell policy PoC on one non-prod agent class with quarantine/deny drills.","unsupportedClaims":["FedRAMP ATO from this announcement","Guaranteed millisecond containment in every environment"]},"engineering":{"fit":"Fit for agent fleets needing runtime sandboxing; Sentry only where BlueField-4 path exists.","architectureIntegration":"Deploy OpenShell gateway/sandbox/supervisor; optionally add Sentry on BlueField-4 as out-of-band monitor.","prerequisites":["OpenShell 0.1.x access","agent identity","policy store","BlueField-4 if Sentry"],"constraints":["Early software maturity","Sentry is reference design","no ATO claimed"],"security":"Enforcement outside agent workload; credentials should not be visible inside the sandbox.","proofOfValue":["Deny unauthorized API write","Quarantine drill logged","Policy change cannot self-approve"]},"delivery":{"work":"Pilot OpenShell policies, logging, and optional Sentry enablement on approved hardware.","dependencies":["Dev resources/GitHub","CPU/DPU inventory","IdP","SIEM sink"],"ownership":"Platform security with AI platform engineering.","skillsAdoption":"Train operators on policy authoring and quarantine runbooks.","governanceCheckpoints":["Policy ownership RACI","Change control for policy prover exceptions","Data-handling review"],"acceptanceCriteria":["Documented deny/quarantine tests","Audit events retained","Rollback path defined"],"risks":["Over-scoping production credentials","Assuming DPU coverage without inventory"]}},"retrievedAt":"2026-09-29T03:02:51Z","enrichedAt":"2026-09-29T03:08:25Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Policy and quarantine runbooks need plain-language paths for non-security developers running agents.","procurementImplications":"Treat as software/reference-design procurement plus optional DPU dependency—not a certified government product claim.","operatingModelImplications":"Establish joint AI-platform and security ownership for agent policy lifecycle.","sourceVerification":{"openedUrl":"https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Launches-Open-Agent-Safety-Platform-to-Secure-Agents-From-Testing-to-Deployment/default.aspx","referenceExcerpt":"Sentry can quarantine agents that attempt to move outside their boundaries in milliseconds","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}