Strategic PartnersIssue 014 records
VAST Data
First VAST Data edition. Higher-ed research computing: Georgia Tech's NSF-funded Nexus opened on an eight-appliance VAST all-flash layer, though it was pre-production and public specs conflict. Agentic AI: VAST and NVIDIA announced an Agent Sandbox Foundation Stack claiming durable agent records, while independent research on the underlying OpenShell runtime shows exfiltration through permitted egress. Sovereign inference: VAST and Hyperfusion plan containerized, air-gapped units, with validation incomplete. No U.S. state/local government VAST evidence surfaced; the CSCS independent evaluation could not be accessed.
- Evidence records
- 4
- Cross-source patterns
- 1
- Evidence classes
- 2 vendor claim1 independent reporting1 independent research
- Outcomes
- 3 emerging1 cautionary
- Source freshness
- 2 new this fortnight1 recent1 older, newly relevant
- Research completed
- 2026-10-30
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Agent-sandbox persistence and audit claims need independent security verification
VAST's new Agent Sandbox Foundation Stack claims durable agent records on top of NVIDIA OpenShell, while independent research on that runtime (Lasso) shows data can leave through policy-permitted egress even when the sandbox behaves as configured. Interpretation: vendor audit and isolation claims should be tested against real attack and reconciliation scenarios before public-sector use.
Operating questionBefore an agency or university pilots sandboxed agents on VAST, how will it prove that every agent action is recorded and that allowed tools cannot exfiltrate data?
Supporting evidenceVAST Data with NVIDIALasso Security
Evidence ledger
Full record · every source keeps its link and limitations
Georgia Tech's NSF-funded Nexus AI supercomputer opens nationally on an eight-appliance VAST Data all-flash storage layer, still pre-production as of Oct 10
Supercomputing News reports that Nexus, a $20M NSF Category I capacity system at Georgia Tech's Coda data center, uses eight VAST Data Ceres V2 all-flash appliances (about 10.8 PB raw NVMe) on a separate storage fabric that is non-blocking for DGX B200 nodes and oversubscribed for RTX PRO 6000 nodes. At least 90% of capacity is allocated nationally through ACCESS/NAIRR peer review. As of October 10, ACCESS listed Nexus as pre-production; the ribbon cutting was October 12. No storage performance results are published.
Why it matters, evidence and limitations
- Why it matters
- Direct higher-education/research relevance: a public university hosting a national NSF resource selected VAST as the shared data platform for mixed AI training, inference and persistent-service workloads. It is a reference deployment for university research-computing (PACE-style) teams, not evidence of measured benefit.
- Evidence and measured results
- Independent trade reporting that cites the Nexus hardware guide, NSF award abstract, ACCESS catalog/operations record and Georgia Tech releases. Facts are configuration and governance (allocation rules, funding category, status); there is no benchmark, baseline or user-outcome data. Separately inspected Nexus portal (operator page) confirms 8x VAST Ceres V2, ~10.8 PB raw, 12.8 TB SCM each, BlueField-3 DPUs, but lists the storage fabric as 200-400 Gb Ethernet and 42 RTX nodes/336 RTX GPUs, versus SCN's NDR InfiniBand storage fabric and 38 nodes/304 GPUs.
- Limitations and uncertainty
- Secondary reporting (AI-assisted drafting per SCN disclosure) with no storage performance data. Operator portal and SCN disagree on storage-fabric type and RTX node/GPU counts; treat specifications as unconfirmed until the hardware guide is checked directly. System was pre-production at reporting time; applications were 'opening soon'. VAST's own social post about the ribbon cutting was not used.
VAST Data announces an Agent Sandbox Foundation Stack with NVIDIA NemoClaw and OpenShell, claiming persistent agent state and durable audit records
VAST announced a Foundation Stack in which NVIDIA OpenShell provides isolated execution and policy enforcement, while the VAST AI OS supplies zero-copy-clone sandbox workspaces (DataStore), shared agent memory/context (DataBase), and a persistent record of agent tool calls, prompts, tokens, costs and decisions, plus VAST Management System observability. VAST says it will demo at GTC Berlin and publish an automated deployment on GitHub. No customer deployment, benchmark or independent validation is cited.
Why it matters, evidence and limitations
- Why it matters
- Limited but timely: state agencies and universities piloting agentic AI need audit trails and data persistence that survive ephemeral agent compute. The announcement is a vendor architecture, not a SLED deployment; relevance is as a design option to evaluate against public-records, audit and security requirements.
- Evidence and measured results
- Vendor press release only (vendor claim). Capabilities are described, not measured. Independent scrutiny of the underlying NVIDIA runtime: Lasso Security (April 2026, this edition) showed policy-permitted exfiltration through default egress, and an independent test already archived on this site (dev.to, Oct 23, 2026) reported OpenShell 0.1.5's documented OCSF audit export produced zero records in two runs.
- Limitations and uncertainty
- Single vendor source with partner quotes; no performance, scale or customer data. GitHub deployment was described as planned. Page date label shows Oct 20 while the dateline is October 21, 2026 (dateline used). Independent findings concern NVIDIA OpenShell, not VAST components; NVIDIA runtime coverage belongs primarily to the NVIDIA stream and is cross-referenced here only because it underpins this stack.
Lasso Security research shows data exfiltration from NVIDIA NemoClaw/OpenShell sandboxes through default policy-permitted egress (npm, node, git/gh)
Lasso demonstrated two attack scenarios in which an agent inside an OpenShell sandbox, using default egress policy, installed a malicious npm package and exfiltrated credentials and files (including plaintext API keys) via allowlisted gh/git paths, with cron-based persistence in the second scenario. Lasso reports NVIDIA's disclosure program deemed the scenarios out of scope. Researchers manually adjusted some default policies (e.g., adding api.openai.com, node path, installing gh) to make the environment functional.
Why it matters, evidence and limitations
- Why it matters
- Newly relevant because VAST's October 21, 2026 Agent Sandbox Foundation Stack builds on NemoClaw/OpenShell; public-sector buyers evaluating that stack need to know sandbox isolation does not stop misuse of permitted egress. Not VAST-specific and not a SLED deployment.
- Evidence and measured results
- Proof-of-concept security research by a security vendor with described attack steps and example packages/PRs; no systematic measurement. Published April 2026, predating the VAST stack; included as independent scrutiny of a component the new stack depends on.
- Limitations and uncertainty
- Tested an early/alpha NemoClaw/OpenShell release in April 2026 with some manual configuration changes; later versions may differ. Lasso sells AI security products (potential interest). Findings concern NVIDIA components, not VAST software; NVIDIA-runtime coverage belongs primarily to the NVIDIA stream.
VAST Data and Hyperfusion plan containerized sovereign AI inference (Sentinel) with VAST AI OS as preferred data platform; technical validation still in progress
VAST and UAE-based Hyperfusion announced plans to integrate the VAST AI OS into Sentinel, a factory-built 45-foot container AI data center (reference 96 GPUs, about 2-3 PB per unit; HyperPod about 20 PB), marketed for rapid, sovereign and optionally air-gapped inference with confidential computing. The release states joint go-to-market will begin while technical validation of VAST within Sentinel and HyperPod is completed.
Why it matters, evidence and limitations
- Why it matters
- Limited: the target market explicitly includes government, and modular, air-gapped, in-jurisdiction inference addresses data-residency and power-availability constraints some states and universities face. There is no U.S. SLED customer, and the offering is not yet validated.
- Evidence and measured results
- Vendor press release with executive quotes (vendor claims). 'Weeks rather than years' and data-residency statements are not substantiated with deployments or measurements; integration is a plan.
- Limitations and uncertainty
- Announcement of intent; technical validation incomplete; no customers named; non-U.S. vendor context limits transferability to U.S. state procurement. Page date label shows Oct 18 while the dateline is October 19, 2026 (dateline used).
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.
- Independent research
- Research conducted outside the implementing organization.