To break down the language barrier, OpenAI is there.
ServiceNow frames it this way. Behind the scenes lies a three-year agreement that will see it, among other things, harness GPT-5.2 to develop technologies for language understanding and vocal synthesis. In the end, we’re promised, there will be multilingual agents that operate without relying on the text modality.
GPT models will also be put to work on the computer use side (“computer user agent”). In particular, the aim is to orchestrate office tools and automate legacy systems (ServiceNow mentions mainframes).
OpenAI Models, Already Deeply Embedded in ServiceNow
In practice, ServiceNow has already established multiple gateways with OpenAI.
On its platform, conversational, generative and agentic capabilities are largely grouped under the Now Assist brand.
The Now Assist products install as plug-ins. They provide access to three types of components: generative skills, agents and agentic flows. These building blocks typically target use within an application. For example, for skills, the generation of documentation on the topic of collaborative work management. For the agentic workflows, obtaining governance guidance in the CMDB.
Some agentic flows operate at the platform level: investigating incidents, creating tasks from images, proposing responses to a survey…
By default, Now Assist relies on an internal service, which combines specialized in-house models* and open models “selected, configured or enhanced by ServiceNow, its community or its partners.”
For certain skills, it is possible to switch to alternative providers: Google (Gemini 2.5 Flash and 2.5 Pro), Anthropic (Claude 3.7 Sonnet on Amazon Bedrock)… or OpenAI (GPT-4.1 and GPT-4.1-mini on Azure OpenAI).
With every application, a “generative AI controller” is installed. It enables interfacing external LLMs via API, at the level of flow designers, agents and scripts. By default, this gives four possibilities:
- Generate text about a ServiceNow topic
- Summarize a topic
- Create a use case and an associated prompt
- Perform sentiment analysis
This controller acts as the gateway with OpenAI and Azure OpenAI. As well as with Google, Aleph Alpha, Amazon (Bedrock) and IBM (watsonx). It also includes a generic connector.
* At the top of the list, a 12B general-use model (flow creation, moderation, code writing…) based on Mistral Nemo Instruct.
