Atlassian and OpenAI expanded their partnership Tuesday in an arrangement that an OpenAI representative described as effectively a spend commitment by Atlassian on OpenAI's models, VentureBeat reported. Atlassian is deploying GPT-6 Astra and the GPT-5.6 series across its products, but its Rovo AI will keep sending work to several model providers rather than defaulting to OpenAI.

What Happened

Both companies posted announcements about the expanded relationship, which builds on a partnership that dates to 2023. Neither side disclosed how much Atlassian has agreed to spend. The OpenAI representative's description makes the direction of the money clear: Atlassian is the buyer, committing dollars to OpenAI's models and technology.

OpenAI's GPT-6 Astra, which shipped Sept. 3 and is priced at $10 per million input tokens and $50 per million output tokens, is being made available across Atlassian's platform along with the GPT-5.6 models. Atlassian confirmed to VentureBeat that GPT-6 Astra will not become Rovo's default model. Instead, an internal AI gateway routes each request across OpenAI and other providers, weighing capability, speed and cost for the task at hand.

RelatedNew York City Comptroller Urges Anthropic to Delay IPO Over Safety and Governance Concerns

The announcement came during Atlassian's Team '26 Europe conference, where the company also detailed a rebuilt Model Context Protocol server. Atlassian first introduced its MCP integration in December 2025. The server now handles more than 15 million calls a day and exposes more than 220 tools across Jira, Confluence, Bitbucket, Loom and Goals, with Jira Service Management support listed as coming soon. In internal testing, Atlassian said, the rebuilt server used up to 25% fewer tokens on comparable Jira and Confluence work.

Why It Matters

The deal shows how enterprise software vendors are buying from model labs without tying their products to one of them. A volume commitment can secure pricing and capacity from OpenAI, while the routing gateway lets Atlassian shift traffic toward cheaper or faster models as tasks and prices change. For customers, that means the model answering a Rovo request may vary, and the commercial terms behind it remain private.

The MCP figures matter because they show Atlassian data flowing to AI tools outside its own products. Scott White, Anthropic's head of enterprise product, called Atlassian's server "one of the most-used enterprise MCP integrations on Claude," according to VentureBeat. That puts Atlassian in the position of serving both OpenAI and its main rival, Anthropic, even as it commits spending to one of them.

Security and Admin Controls

Opening Jira and Confluence to outside agents raises questions about who can read or change what. Atlassian said the MCP server uses OAuth 2.1 with PKCE by default and checks each user's permissions at query time, so an agent sees only what the person behind it is allowed to see. Actions are split into read, write and destructive tiers.

Administrators can apply data security policies and data loss prevention through Atlassian Guard, restrict access with domain and IP allowlists, review audit logs and switch off write actions entirely. Those controls are aimed at large customers that want AI assistants working inside project and documentation systems without granting broad access.

What to Watch

Atlassian has not said how large the OpenAI commitment is or how much Rovo traffic will move to OpenAI models as a result. VentureBeat characterized the deal as a deepening of an existing relationship rather than an exclusive arrangement.

The arrival of Jira Service Management tools on the MCP server would extend outside AI access into IT support and service desk workflows. Customers will also be watching whether the reported token savings hold up outside Atlassian's internal tests, since token use drives the cost of running AI agents against large Jira and Confluence instances.