{"resourceId":"deepset-self-improving-sovereign-agents-reference-architecture-nvidia-20261021","versions":[{"version":"external-cfbee6838ab01d5d0d5fb0621354cbd5752051d0ad207b72b243c49b83cd03c7","resource":{"id":"deepset-self-improving-sovereign-agents-reference-architecture-nvidia-20261021","title":"deepset and NVIDIA publish a reference architecture for self-improving agents in air-gapped environments, with proposed changes gated by evaluation, approval, audit and rollback","organization":"deepset; NVIDIA","sector":"AI agent software (governments and regulated organisations)","geography":"Europe launch (GTC Berlin); global","publishedAt":"2026-10-21","publicationDate":"2026-10-21","eventDate":"2026-10-22","sourceName":"PR Newswire (deepset)","sourceLabel":"Vendor press release","sourceUrl":"https://www.prnewswire.com/news-releases/deepset-unveils-reference-architecture-for-self-improving-sovereign-ai-agents-on-nvidia-infrastructure-302914982.html","evidenceClass":"vendor-claim","outcomeClass":"emerging","topics":["developers-agents","data-security","governance-procurement"],"finding":"deepset, whose Haystack platform underpins the European Commission's AI@EC, released an open reference architecture with NVIDIA for self-improving agents in locally controlled and air-gapped environments. One Haystack agent performs the task; a second analyses traces and proposes changes to the agent harness, each evaluated against predefined objectives and the current version before promotion, with approval, audit and rollback controls. Components are Nemotron models via NIM, NeMo Gym for optimisation, and NemoClaw blueprints running the agent inside OpenShell with policy controls over files, networks, credentials and tools. The first demonstration is a cybersecurity agent analysing confidential source code; a companion GitHub repository and a DGX Spark log-analysis walkthrough are provided.","sledRelevance":"Relevant to state and university security operations and to any agency considering agents that change their own configuration: it names concrete control points (evaluation gates, approval, rollback) that a U.S. public buyer would need before allowing an agent to self-modify. No public-sector deployment is reported.","evidence":"Vendor release describing an architecture and demo; no measured improvement, accuracy or false-positive data. The 'self-improving' loop is described, not evaluated. Related OpenShell containment limits were archived Oct 21 (configuration-dependent).","architectureImplications":"Two-agent loop (task agent plus optimiser agent) with an evaluation environment; runs on local NVIDIA hardware down to DGX Spark; OpenShell for isolation.","governanceImplications":"Makes change management of agent behaviour explicit: proposed harness changes need evaluation and human approval before production.","securityPrivacyImplications":"Keeps confidential code and logs local; agent permissions enforced by OpenShell policy, whose effectiveness depends on operator configuration.","caveats":"Vendor claims; demo only; no independent evaluation; optimisation objectives could reward the wrong behaviour if poorly specified.","streamIds":["nvidia"],"roles":{"sales":"Sales framing: the pitch is agents that improve themselves without leaving the building, under human approval. Interpretation: for a state security operations centre or university IT security team, the credible offer is a governed pilot on a bounded task such as code or log review, not autonomous improvement. There are no measured results, so value claims must stay at the level of control design and data locality.","engineering":"Engineering framing: the architecture pairs a task agent with an optimiser agent, gates proposed harness changes through evaluation against objectives, and runs execution inside OpenShell. Pre-sales engineers should check that evaluation datasets reflect the customer's real cases, that approval and rollback are wired into existing change management, and that OpenShell policies are restrictive by default. Air-gapped operation needs local model updates and artifact mirroring.","delivery":"Delivery framing: self-improvement adds a release pipeline for agent behaviour. Delivery teams should define who approves changes, how evaluation results are recorded, and how rollback is triggered, and treat each promoted change like a software release. Start with a read-only task. Acceptance criteria below are proposals; the release reports a demonstration, not observed outcomes. Keep a record of every promoted change, its evaluation scores and its approver, so auditors can reconstruct agent behaviour later."},"roleDetails":{"sales":{"customerProblem":"Security teams need help reviewing sensitive code and logs without sending them to external AI services.","stakeholders":["State or campus CISO","Security operations lead","Change advisory board"],"discoveryQuestions":["Which bounded, read-only task would you pilot?","Who would approve agent configuration changes?","Can you operate air-gapped model updates?"],"valueHypothesis":"Interpretation: a locally run, governed agent could assist analysts on code or log review; benefit unmeasured.","potentialEngagement":"Governed pilot design for a local security-analysis agent with approval and rollback controls.","unsupportedClaims":["Measured accuracy or analyst time savings","Safe autonomous self-modification","Production public-sector use"]},"engineering":{"fit":"Organisations with confidential code or logs and local GPU capacity.","architectureIntegration":"Haystack task and optimiser agents, Nemotron via NIM, NeMo Gym evaluation, NemoClaw in OpenShell on local NVIDIA hardware.","prerequisites":["Local GPU capacity","Evaluation dataset of real cases","Change-management integration"],"constraints":["Demo-stage","Objectives must be well specified","NVIDIA runtime stack dependency"],"security":"Enforce least-privilege OpenShell policies and log all agent actions and promoted changes.","proofOfValue":["Compare agent findings with analyst baseline on historical cases","Test rollback of a promoted change","Red-team the sandbox policy"]},"delivery":{"work":"Deploy locally, define evaluation and approval flow, pilot on a read-only task.","dependencies":["Hardware","Security team time","Evaluation data"],"ownership":"Security operations, with change advisory board approval.","skillsAdoption":"Analysts need training to review agent findings and proposed changes.","governanceCheckpoints":["Policy review before pilot","Approval for each promoted change","Monthly evaluation review"],"acceptanceCriteria":["Proposed: precision and recall against analyst-labelled cases","Proposed: every promoted change has recorded evaluation and approver","Proposed: rollback executed within agreed time in test"],"risks":["Optimiser reinforcing wrong behaviour","Over-permissive sandbox","Reviewer fatigue"]}},"retrievedAt":"2026-10-23T02:58:00Z","enrichedAt":"2026-10-23T03:07:07Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Shifts analyst work toward reviewing agent-proposed changes; requires staff able to judge evaluation results.","procurementImplications":"Open-source components reduce licence lock-in for orchestration, but the model-serving and runtime stack is NVIDIA's.","operatingModelImplications":"Introduces a release process for agent harness changes similar to software change control.","sourceVerification":{"openedUrl":"https://www.prnewswire.com/news-releases/deepset-unveils-reference-architecture-for-self-improving-sovereign-ai-agents-on-nvidia-infrastructure-302914982.html","referenceExcerpt":"with approval, audit and rollback controls keeping people in charge of what reaches production","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}