OpenAI plugged ChatGPT into Oracle and ServiceNow in two days. Only one plugin is actually live

In two days OpenAI expanded its partnerships with two large enterprise software vendors. Oracle and ServiceNow both want employees to work with their systems directly from ChatGPT, and both are building security products on OpenAI's cyber models.

ServiceNow shipped its ChatGPT plugin now and pushed the rest into 2027

On October 28 ServiceNow made its MCP plugin available in the OpenAI marketplace. A company connects ChatGPT to its own ServiceNow instance and chooses which capabilities to expose. According to OpenAI, an IT team member can then find similar incidents, assess the risk of a planned change and update a record in a single conversation. ServiceNow Action Fabric enforces the existing business rules and permissions, so a user can do in ChatGPT only what they could do in ServiceNow itself. ServiceNow says actions are still executed and governed in its own system.

The rest of the deal is a plan for now. Enhanced plugin capabilities are expected by the end of 2026. Voice-driven risk questionnaires for AI projects in the Super Intelligence Control Tower are slated for the first quarter of 2027. The joint security solution is due for early access in January 2027 and general availability in the second quarter of 2027. Codex in AI Workflow Factory should arrive during 2027. ServiceNow will also give ChatGPT Work and Codex to all of its employees, not only engineers, and OpenAI itself uses ServiceNow workflows in its operations.

A day earlier, on October 27, Oracle announced Oracle AI Workspace. It is meant to bring Oracle Fusion Applications, NetSuite, Oracle AI Database and OCI into ChatGPT through plugins. OpenAI describes it as still in development. Oracle also introduced Oracle AI Security Mesh, built on OpenAI cyber models, App Composer for building apps with Codex, and a small business bundle with $12,000 a year in OCI credits. According to OpenAI, Oracle reports 130,000 active ChatGPT users and more than 95,000 active Codex users.

ChatGPT becomes the entry point to enterprise systems while vendors keep the rules

Both vendors chose the same division of labor. ChatGPT is where a person asks and assigns work. Oracle and ServiceNow keep the data, the approvals and the permissions. Atlassian made a similar move in early October, which Radar covered. Within a single month OpenAI announced three expanded partnerships that turn ChatGPT into a shared entry point to the systems where companies track tickets, invoices and projects.

For IT departments this has a practical consequence. The plugin inherits whatever permissions a company already has in ServiceNow or Oracle. If those are a mess, with stale roles or overly broad access, ChatGPT will not cover it up. It simply opens a more convenient and faster path to it. Cleaning up roles becomes a precondition for rollout rather than optional housekeeping.

The second common thread is security. ServiceNow says a single codebase scan can return 10,000 to 100,000 or more findings, and the hard part is telling which ones are actually on fire. The company recently joined OpenAI's Trusted Access for Cyber program. Oracle promises to run the testing infrastructure and the safeguards against outages during tests itself. Cyber models will therefore reach companies through a vendor that carries the operational risk.

Customers can try one plugin today and the rest is a calendar of promises

Out of the whole package, a ServiceNow customer can try only the basic plugin today. The security solution, voice and Codex in AI Workflow Factory are dated 2027, and Oracle AI Workspace has no date at all. According to OpenAI, buying the OpenAI API, Codex and ChatGPT Work through Oracle Marketplace is aimed at customers in the U.S. European companies will have to handle licensing the usual way.

The announcements are also silent on the audit trail. When ChatGPT changes a record at a user's request, ServiceNow executes the action under its own rules. What the announcements do not explain is how precisely the system logs what the model proposed and what the human confirmed. For regulated companies that will be one of the first questions in procurement.

January early access will show whether the models can triage findings

The first hard test is ServiceNow's security solution in January 2027. If customers get a short list of genuinely exploitable bugs out of tens of thousands of findings, the deal has substance. If they get yet another ticket queue, they have gained one more dashboard.

The second signal is Oracle AI Workspace. Until it has a date and a price list, it remains an intention. The third signal is how many more vendors plug into ChatGPT on the same model. The more of them there are, the more ChatGPT becomes the main corporate desktop and the stronger OpenAI's hand at the negotiating table.

Lilith's verdict

Oracle and ServiceNow lent ChatGPT the remote control to their systems, but wired the confirm button into their own rules. Until January 2027 the IT technician holding it mostly flips channels: find a similar incident, update the record.