ServiceNow in ChatGPT: check what can act now and what comes in 2027

Timeline separating the October ServiceNow plugin listing from expected year-end additions and planned 2027 capabilities. Original editorial illustration by PING! Magazine. Sources: OpenAI and ServiceNow.

The practical question in OpenAI and ServiceNow's expanded partnership is what a conversation can change in the workplace. The 28 October announcement puts a ServiceNow MCP plugin in the OpenAI marketplace, while much of the wider programme remains on a timetable. A listing is a starting point for an administrator, rather than evidence that a particular company has enabled it.

OpenAI describes a connection to the customer's own ServiceNow instance, with capabilities chosen by the customer and existing rules applied through ServiceNow's Action Fabric. Its example combines incident research with updating a record. PING has not connected an instance or tested those permissions. Announcement and availability

Read the timetable by feature

OpenAI says enhanced plugin capabilities are expected by the end of 2026. Voice features are expected in the first quarter of 2027; security early access is expected in January, with general availability planned for the second quarter. Codex in AI Workflow Factory is expected during 2027. These are the supplier's plans, not delivered features that PING has verified.

That distinction matters when preparing a pilot or a budget. Write down the exact feature being evaluated and its present release state. A demonstration of one workflow should not become a purchase assumption about another. The collaboration itself also predates this week: ServiceNow published an earlier agreement on 20 January. January announcement

Begin with one ordinary record

Our suggested evaluation starts in a separate demonstration environment with invented records. Ask an authorised user to find one incident, identify where the answer came from and make one permitted change. Repeat with a user who lacks that permission. Keep the expected result in writing before running the exercise, so an attractive answer cannot redefine what success meant.

Also test a failure: an unavailable connection, an ambiguous request or an attempt to change the wrong record. The team should know who sees the failure, whether anything was saved and how the normal ServiceNow interface can be used to inspect the result. These are proposed acceptance checks, not experiments PING has performed or claims that the plugin provides particular audit controls.

Ask the local deployment questions

For a team working across London, Lagos or Nairobi, geography should enter the administrator's checklist rather than the headline as an assumed rollout. Ask which accounts, subscriptions, regions and data arrangements apply to that organisation. Identify who owns the connected instance and who approves a change. This announcement does not establish those answers for any African or UK employer.

A useful first review can therefore fit on one page: the feature and its release state, the records it can read, the actions each role can take, the expected failure behaviour and the person who signs off the pilot. The payoff is a clearer workplace process that can be inspected by its owner. More source-led explanations are on PING's AI desk.