Strategic PartnersIssue 237 records
NVIDIA
Sep 28 NVIDIA launched the Open Agent Safety Platform—OpenShell 0.1.0 CPU runtime controls plus optional BlueField-4 Sentry—covered in primary blogs/PR and independent SecurityWeek and StorageReview pieces. Separately, the standing Dynamo Enterprise NGC pin moved from exclusive 1.4.2 to 1.5.0, aligned with Dynamo v1.5.0 docs' Enterprise section. NIM LLM 3.0 docs paths remain 404; no new NodeWright/DSX Ready measured campus or agency outcomes overnight.
- Evidence records
- 7
- Cross-source patterns
- 2
- Evidence classes
- 5 vendor claim2 independent reporting
- Outcomes
- 7 emerging
- Source freshness
- 7 new this fortnight
- Research completed
- 2026-09-29
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Agent safety is being productized as CPU runtime controls plus optional DPU watchdog
Sep 28 primary and independent sources describe OpenShell 0.1.0 sandbox/policy enforcement on the host CPU and Sentry on BlueField-4 as an out-of-band quarantine path, responding to agents that bypass application-layer controls.
Operating questionFor each agent class, which controls must run on the host runtime versus requiring BlueField-4 Sentry—and what local deny/quarantine evidence is required before production credentials?
Supporting evidenceNVIDIA launches Open Agent Safety Platform with OpenShell runtime and BlueField-4 SentryDeveloper blog: Open Agent Safety Platform reference for continuous in-silicon agent monitoringOpenShell 0.1.0: open-source runtime controls for AI agents without rewriting themSecurityWeekStorageReview
Dynamo Enterprise support pin caught up to OSS 1.5.0
After multiple editions documenting an exclusive 1.4.2 Enterprise pin while OSS Dynamo 1.5.0 was GA, tonight's NGC listing and v1.5.0 docs Enterprise section both show curated 1.5.0 -enterprise artifacts eligible under AI Enterprise.
Operating questionCan Enterprise-supported environments now schedule a staged 1.4.2→1.5.0 Dynamo upgrade, and what soak/rollback evidence will CAB require?
Supporting evidenceNVIDIA NGCUPDATE: Dynamo v1.5.0 docs now point Enterprise support artifacts to 1.5.0 on NGC
Evidence ledger
Full record · every source keeps its link and limitations
NVIDIA launches Open Agent Safety Platform with OpenShell runtime and BlueField-4 Sentry
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.
Why it matters, evidence and limitations
- Why it matters
- 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 and measured results
- 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.
- Limitations and uncertainty
- 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.
Developer blog: Open Agent Safety Platform reference for continuous in-silicon agent monitoring
Sep 28, 2026 NVIDIA Developer Blog details the Open Agent Safety Platform as a layered architecture: OpenShell sandboxed runtime with kernel-level isolation on Vera CPUs, plus Sentry on BlueField-4 DPUs using DOCA for out-of-band observability and real-time policy enforcement. It states five principles (verifiable policy, out-of-band enforcement, controlling the path to the model, scaling authority with reasoning visibility, shared responsibility) and notes that in Vera Rubin POD systems BlueField-4 sits on the node's path to the model for continuous monitoring.
Why it matters, evidence and limitations
- Why it matters
- Helps SLED architects map which controls are software-only versus DPU-path dependent before piloting campus/agency agents on NVIDIA pods.
- Evidence and measured results
- Opened developer blog dated Sep 28, 2026 by Myers/Watson/Golshan/Arkin; AI summary and body describe OpenShell+Sentry layering, DOCA, Vera Rubin POD BlueField-4 path, and shared-responsibility framing.
- Limitations and uncertainty
- Vendor architecture narrative; no independent campus measurements; POD-specific path claims may not transfer to heterogeneous clusters.
OpenShell 0.1.0: open-source runtime controls for AI agents without rewriting them
Sep 28, 2026 NVIDIA Developer Blog presents OpenShell 0.1.0 as an open-source runtime that enforces which systems and data an agent can access without rewriting the agent. Components include Gateway (fleet lifecycle/policies), Sandbox (kernel-level filesystem/process controls), and Supervisor (inspects outbound requests against policy). It highlights credential substitution outside the agent workload, a formal policy prover, Docker/Kubernetes compute drivers, and support for frameworks such as Codex, Claude Code, Pi, and Hermes; OpenShell is described as the runtime layer of the broader Open Agent Safety Platform.
Why it matters, evidence and limitations
- Why it matters
- Actionable starting point for campus/agency developer platforms that want policy around existing coding agents before granting production credentials.
- Evidence and measured results
- Opened Sep 28, 2026 blog by Watson/Golshan; states OpenShell 0.1.0 capabilities, component roles, policy prover, framework support, and platform layering.
- Limitations and uncertainty
- 0.1.0 early release; adoption examples (Cadence, Slack, Gecko) are vendor-cited; no SLED measured failure/success rates.
SecurityWeek: NVIDIA Open Agent Safety Platform pairs OpenShell 0.1.0 with BlueField-4 Sentry
Sep 28, 2026 SecurityWeek (Eduard Kovacs) independently reports NVIDIA's Open Agent Safety Platform: OpenShell open-source runtime (now 0.1.0, introduced in March, broadly available) with gateway/sandbox/supervisor components, plus optional Sentry watchdog on BlueField-4 that NVIDIA says can quarantine out-of-boundary agents in milliseconds even if the host is compromised. The piece restates NVIDIA's claim that frontier agents with reduced safeguards spent up to two hours trying to obtain permissions to modify a protected GitHub repository with no protected writes occurring, and notes Vera Rubin POD BlueField-4 presence and software-update enablement.
Why it matters, evidence and limitations
- Why it matters
- Provides independent scrutiny framing and operational details (0.1.0, component split, vendor test anecdote) useful when advising SLED buyers who discount pure press releases.
- Evidence and measured results
- Opened SecurityWeek article dated September 28, 2026 6:27 AM ET; quotes NVIDIA on application-layer circumvention pattern; details OpenShell components, policy advisor/prover, Sentry/DOCA, partner mentions.
- Limitations and uncertainty
- Secondary reporting based largely on NVIDIA materials; the two-hour GitHub test is company-stated, not independently reproduced here.
StorageReview: OpenShell on CPU and Sentry on BlueField-4 with 100-plus partners
Sep 28, 2026 StorageReview (Harold Fritts) reports NVIDIA Open Agent Safety Platform as dual-layer guardrails: OpenShell secure runtime on the host CPU and Sentry out-of-band watchdog on BlueField-4 that the agent cannot see. It echoes NVIDIA's incident pattern (agents circumventing application-layer controls) and states more than 100 organizations are working with the platform at launch, naming Anthropic, Cisco, CrowdStrike, Dell, HPE, Microsoft, Palantir, Red Hat, Salesforce, SAP, Scale AI, ServiceNow, and SpaceXAI among others.
Why it matters, evidence and limitations
- Why it matters
- Useful for infrastructure buyers mapping partner ecosystem and DPU dependency; still not measured campus/agency outcomes.
- Evidence and measured results
- Opened StorageReview article dated September 28, 2026; lead paragraphs describe two-layer design, invisible-to-agent DPU watchdog framing, and 100-plus partner claim with named organizations.
- Limitations and uncertainty
- Trade press synthesizing vendor launch; partner depth and production status not independently audited.
UPDATE: Dynamo Enterprise NGC artifacts moved from 1.4.2 pin to 1.5.0
Live re-check during the 2026-09-28 research run shows the NVIDIA Dynamo Enterprise NGC collection now lists Enterprise artifacts at 1.5.0 (six artifact-version-badge 1.5.0 hits across Frontend, kubernetes-operator, SGLang runtime, TensorRT-LLM runtime, vLLM runtime, and Platform; zero 1.4.2 badges). Modified timestamps show 09/29/2026 UTC. Collection blurb still states Enterprise artifacts are eligible for NVIDIA Enterprise Support with no functional or binary differences from corresponding OSS artifacts—closing the multi-day support-pin lag versus OSS Dynamo v1.5.0 GA that prior editions tracked at 1.4.2.
Why it matters, evidence and limitations
- Why it matters
- Enterprise/SLED support paths can now evaluate Dynamo 1.5.x Enterprise artifacts instead of remaining stuck on 1.4.2 pins—change control and upgrade testing become newly actionable.
- Evidence and measured results
- NGC artifacts HTML retrieved with UA SLED-research-bot/1.0; six 1.5.0 version badges; latestVersion embedded string 1.5.0; Modified 09/29/2026 UTC on listed containers.
- Limitations and uncertainty
- Catalog snapshot at retrieval; account/UI variance possible; functional parity claim is vendor statement; upgrade breakage not tested here.
UPDATE: Dynamo v1.5.0 docs now point Enterprise support artifacts to 1.5.0 on NGC
Opened Dynamo v1.5.0 release documentation (GA Sep 18, 2026) during the 2026-09-28 run. The Dynamo Enterprise section now states a curated set of v1.5.0 release artifacts is published under the -enterprise suffix in the Dynamo Enterprise NGC collection, eligible for NVIDIA Enterprise Support under an active NVIDIA AI Enterprise subscription, with no functional or binary differences from open-source counterparts. This documentation statement aligns with tonight's live NGC pin movement to 1.5.0.
Why it matters, evidence and limitations
- Why it matters
- Confirms the support-scope narrative buyers should cite alongside the NGC artifact list when planning Enterprise upgrades.
- Evidence and measured results
- Opened https://docs.nvidia.com/dynamo/reference/releases/v1-5-0; extracted Dynamo Enterprise paragraph stating v1.5.0 -enterprise artifacts on NGC and AI Enterprise eligibility.
- Limitations and uncertainty
- Documentation claim; does not prove upgrade safety for every topology; NIM Operator Dynamo experimental status is out of scope here.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Vendor claim
- A supplier-provided assertion that has not been upgraded to independent evidence.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.