Technology and Scienceupdated 52m ago2 min read
Nvidia Launches Nemotron 3.5 Super VL and Bolt Open-Weight Models for Enterprise Agents
This article is an automated aggregation of coverage from news outlets. Our editorials offer in-depth analysis curated by the NewsCord team.
Nemotron 3.5 Super VL is a 120-billion-parameter enterprise-grade model for AI agents. Bolt models enable local PC deployment with 74-billion parameters and 24GB memory.
Nvidia opens agent models
Nvidia launched new open-weight models for agents in enterprises and on PCs, including the Nemotron 3.5 Super VL with 120 billion parameters and two Bolt models intended to run locally. Nvidia said the Super VL model is designed for agents in the business domain and can complete tasks faster than competing Anthropic and OpenAI models in post-training-specific tasks such as document intelligence, financial accounting, scientific experimentation, medical insurance claims, and resilience to security attacks.
Nvidia also released Bolt models for local deployment on AI PCs using the Nvidia RTX Spark chip and on workstations with the DGX Spark, with Nemotron 3.5 Bolt 74B-A9B 24GB aimed at local tasks and Bolt E4B aimed at conversational agents. Nvidia Vice President of Open Models and Software Kari Briski said in a statement, “any organization that wants to run agents on Nvidia’s infrastructure now has an open model that is optimized for its level without having to choose between capacity and deployability.”
Security shifts to enterprises
Microsoft and Nvidia urged enterprises to assess the security of open-weight AI models throughout deployment, warning that changes to model weights and configurations can weaken safeguards even when a model retains its ability to perform assigned tasks. The companies’ joint paper identified 30 risk types across five categories and advised enterprises to verify model provenance, test configurations, and reassess risks after material changes, while stating, “Weight openness alone does not determine an AI system’s risk.”
Michael Scovetta, Partner Security Assurance Manager at Microsoft, said, “Weight openness alone does not determine an AI system’s risk,” and he added that “Open- and closed-weight models can share many of the same underlying risks, but how those risks materialize depends on the deployment.” The guidance also warned that open-weight models can be fine-tuned, merged, modified, or repackaged, and it cautioned that “Safety behaviors are often the more fragile of the two,” as changes can degrade safeguards while leaving task performance intact.
Provenance, monitoring, and controls
Microsoft and Nvidia recommended using cryptographic signatures to verify that a model matches the publisher’s original artifact, and they advised that “No volume of behavioral testing establishes what an artifact is, only how it behaves.” The guidance said that where signatures are unavailable, organizations should record the unverified integrity as an open risk rather than treating evaluation results as closing it, and it urged monitoring model inputs, outputs, and tool use after deployment.
Gartner recommended routing open-weight model traffic through governed AI gateways to enforce authentication, prompt-injection defenses, sensitive-data protection and audit logging, and it also recommended isolating self-hosted runtimes and applying controls such as private network paths and customer-managed encryption keys for cloud API consumption. Sanchit Vir Gogia, chief analyst at Greyhound Research, said, “Self-hosting is a staffing decision as much as an infrastructure one, and the people to carry it must move with the model,” while warning that self-hosting does not necessarily replace all safeguards available through managed services.