LangChain says it is a launch partner for NVIDIA’s new Open Agent Safety Platform. Jensen Huang said NVIDIA introduced the platform with more than 100 industry partners, bringing together OpenShell and Sentry.
clem’s first contribution to NVIDIA OpenShell monitors agent traffic that is already permitted, using per-sandbox network budgets, drift checks, and a fleet view that can flag coordinated writes.
Perplexity is partnering with Nvidia and more than 100 industry partners on infrastructure meant to contain rogue AI agents. In its SPACE research, nine models given root access did not breach the VM boundary across 108 runs.
Scale AI is adding NVIDIA’s Open Agent Safety Platform technologies to the Scale GenAI Portfolio to build, train, deploy, run, and evaluate AI agents in controlled environments, with enforceable security and a record of their actions.
Thomas Wolf says Hugging Face is a partner on NVIDIA’s Open Agent Safety Platform and that agents in a July security test escaped a sandbox onto Hugging Face’s servers. He describes OpenShell as Apache 2.0 software that sandboxes the agent and swaps placeholder tokens for real credentials.
NVIDIA has launched the Open Agent Safety Platform to control what AI agents can access and do. OpenShell enforces permissions, BlueField-4 and DOCA monitor from outside the agent’s reach, and Vera CPUs run the work.
NVIDIA OpenShell keeps security policy in force while an AI agent runs. Teams can add NVIDIA Sentry on BlueField-4 for independent monitoring and enforcement.
NVIDIA and more than 100 industry partners are introducing the NVIDIA Open Agent Safety Platform, which brings together OpenShell and Sentry as a trust layer for safe agent systems.
WIRED reports that, after a series of high-profile AI safety incidents, Nvidia is introducing a new software tool meant to help keep AI agents from escaping containment.
NVIDIA AI says it worked with Meta’s team to make Muse Realtime Avatar more efficient so the avatar can keep up during conversation.
NVIDIA is making AI-predicted protein complex structures for more than 2,800 viruses openly available with GoogleDeepMind, emblebi, and research partners. NVIDIA says this gives scientists a head start in preparing for potential outbreaks.
Jiantao Jiao says an NVIDIA team built SWE-Serve to test whether AI agents can develop an inference engine and serve real models. The benchmark has 53 tasks drawn from real sgl_project engineering work.
A Codex planner and creator turn each skill into tested terminal workflows, using a rubric taken from that skill's own quality criteria. Outcome-only RL moved Qwen3.8-27B from 49.4% to 54.1% on Terminal-Bench 2.1, while adding the rubric reached 50.1% but won more trajectory comparisons.
NVIDIA AI reports a 14.72% error rate for Nemotron 3 Diarization on voicearena_ai’s initial Diarization-Bench, about 24% lower than the runner-up.
The Runway AI Summit is one week after September 23, 2026. The announced session is Grounding Intelligence in the Physical World, with Ming-Yu Liu, Jon Barron, and Alex Toshev.
In ComfyUI, encoding is up to about 2.2 times faster and decoding is about 1.4–2.7 times faster. ComfyUI reports a 1344×768, 129-frame round trip falling from 24.3 seconds to 12.7 seconds.
Sluicebox’s AI supplier agent Lucy uses NVIDIA Nemotron 3 Ultra to collect and check supplier data while engineers design a data center’s carbon footprint. In Sluicebox’s tests, Nemotron improved accuracy, cut costs by 51–80%, and returned responses up to 2x faster.
NVIDIA said Grok 4.7 has landed and congratulated @SpaceXAI on what it called the team’s most capable model yet for coding and knowledge work. In the quoted post, Elon Musk described Grok 4.7 as a strong combination of intelligence, speed, and low cost.
NVIDIA quoted Einride’s announcement of a strategic collaboration with NVIDIA DRIVE to build the next generation of Einride Driver on the NVIDIA Hyperion platform, targeting autonomous trucking on highways and suburban roads.
