OpenAI widens ServiceNow deal; Nvidia ships speech model

OpenAI and ServiceNow say they are expanding their partnership so OpenAI models and agents can act inside ServiceNow’s business software. The first piece, a ServiceNow plugin for ChatGPT, is available now, the companies say.

The expansion builds on a partnership announced in January, according to an OpenAI post and a ServiceNow release, both Wednesday. Neither gives a dollar figure.

What the companies say is coming

With the plugin, an IT worker can find similar incidents, assess the risk of a change and update a ServiceNow record without leaving ChatGPT, the companies say. Customers choose which capabilities to turn on. ServiceNow says its Action Fabric layer applies the customer’s existing business rules and permissions. It says ServiceNow remains the system where those actions run.

The companies plan a Codex integration in ServiceNow’s AI Workflow Factory for building apps. OpenAI models will also power two ServiceNow development tools, Build Agent and Autonomous Engineer, the post says.

They also plan a joint security product. ServiceNow says a codebase scan can turn up 10,000 to 100,000 existing and synthetic vulnerabilities or more, and the product is meant to help teams sort which ones to fix first. ServiceNow says it recently joined OpenAI’s Trusted Access for Cyber program.

A voice feature in ServiceNow’s Super Intelligence Control Tower would let a risk professional describe an AI use case aloud and have an agent fill in assessment fields, ServiceNow says.

ServiceNow plans to give all its employees ChatGPT Work and Codex. OpenAI says it uses ServiceNow in its own operations. OpenAI is also a founding member of ServiceNow’s new Control Tower Alliance, the release says.

“Together, we are unlocking extraordinary value for our customers by turning world-class reasoning into trusted execution at global scale,” ServiceNow Chairman and CEO Bill McDermott said in the release.

The companies say voice is expected in the first quarter of 2027, security early access in January 2027 with general availability planned for the second quarter, and Codex in AI Workflow Factory sometime in 2027. ServiceNow’s release says actual results could differ.

Our read: The plugin is the part that exists today, and it matters more than the road map. It lets a chatbot change records in a system companies use to run IT and security. The safeguard both companies point to is that ChatGPT inherits ServiceNow’s permissions, so an employee can do only what the employee could already do. That is a sound design, but it also means a misconfigured permission now has a chat window in front of it. The security product addresses a real problem, since scanners already find more flaws than teams can fix. It is not due until 2027, so for now it is a promise. Neither company disclosed the money involved, and the release’s own caveats say plans could change. The test is what ships, not the list.

Nvidia releases an open transcription model

Nvidia released Nemotron 3.5 Transcribe, an open-weight model that turns recorded speech into text, according to a Hugging Face post by Maryam Motamedi and Nvidia’s research page. Nvidia says it transcribes 40 languages, translates 31 of them to or from English and labels speakers in 13. It handles recordings up to an hour long and up to eight speakers, Nvidia says.

The model has about 30 billion parameters, with about 4 billion active per token, the post says. Nvidia’s Hugging Face post says it ranks first among open-weight models on Hugging Face’s open speech-recognition leaderboard for English, at a 5.33% average word error rate. It is built for recorded audio, not live streaming.

Nvidia also released Granary v2, the training data pipeline behind the model. The public manifest covers 463,062 hours of speech in 48 languages, up from 25 in the first version, Nvidia says. In a medical case study, Nvidia says fine-tuning cut word error rate on held-out speakers by 45.8% relative to the best prompt.

What’s next

ServiceNow says enhanced plugin features are expected by the end of 2026. Nvidia’s Hugging Face post says quantized versions of Transcribe for local use are coming soon.

Sources: OpenAI, ServiceNow, Nvidia on Hugging Face, Nvidia Research