Atlassian deepens its partnership with OpenAI, integrating newer GPT-6 Astra and GPT-5.6 models into its enterprise software while maintaining a multi-provider strategy and strengthening data governance measures.
Atlassian has deepened its relationship with OpenAI, bringing the newer GPT-6 Astra and GPT-5.6 models into its enterprise software stack while stopping well short of tying the platform to a single supplier. According to Atlassian and OpenAI, the move extends a partnership that began in 2023 and is intended to improve how teams plan, build and deliver work across Jira, Confluence, Bitbucket and Rovo AI. The companies say the collaboration is built around Atlassian’s Teamwork Graph, which links people, projects, documents and decisions across the organisation.
The practical effect is less a wholesale redesign than a broader set of connectors. Atlassian has long used OpenAI models inside Atlassian Intelligence, and it already introduced an MCP connection for ChatGPT in late 2025. This week’s changes centre on deeper integration: newer OpenAI models are being added to Rovo’s mix as they are released, and Atlassian has rebuilt its Model Context Protocol server to expose more of its enterprise data to external AI tools. OpenAI described the commercial element of the arrangement as, in effect, a spend commitment from Atlassian.
For buyers, the most notable detail is what has not changed. Atlassian has said GPT-6 Astra is not becoming the default model for Rovo. Instead, Rovo continues to route work through an internal gateway that can draw on OpenAI and other providers, balancing quality, latency and cost for each task. That model-agnostic approach is reinforced by Atlassian’s own materials, which highlight its work with Anthropic as well as OpenAI, suggesting that the company is treating models as interchangeable components rather than a single strategic dependency.
The centrepiece of the announcement is the expanded MCP layer. Atlassian says its rebuilt server now handles more than 15 million calls a day and supports more than 220 tools across Jira, Confluence, Bitbucket, Loom, Goals and related products, with Jira Service Management support due later. The company says the new version uses fewer tokens than its predecessor in internal testing, and that it allows an agent in ChatGPT, Codex, Cursor or Claude to move across the software lifecycle in one request rather than stopping at a simple search or retrieval step.
Atlassian has also made the governance model more explicit, which will matter to security teams. The company says the MCP server is hosted by Atlassian, uses OAuth 2.1 and PKCE, and respects each user’s existing permissions at query time. It also separates read, write and destructive actions into different risk tiers, allows administrators to set allowlists and audit logging, and supports controls through Atlassian Guard to enforce data security policies and data loss prevention. Those measures do not eliminate the risk of prompt injection, but they do provide a clearer framework for testing autonomous workflows against enterprise controls.
OpenAI’s GPT-6 Astra brings additional capability, but also higher cost and more caution. OpenAI says the model has set results on computer-use, coding and cybersecurity benchmarks, is available to a broad set of ChatGPT users and through the API, and costs $10 per million input tokens and $50 per million output tokens. The company also says the model can trigger safety interventions that pause or halt work, including in security-sensitive use cases, and that some of its reasoning is harder to monitor than earlier versions. For enterprises, that means any serious deployment will need to be designed to tolerate interruptions and to rely on system logs as much as on the model’s own explanations.
The data picture is relatively clear, but still worth scrutiny. Atlassian says only the specific data returned by an MCP tool is passed to OpenAI as context for that request, and only within the boundaries of the requesting user’s permissions. OpenAI says GPT-6 Astra supports zero data retention for eligible API customers, while Atlassian says OpenAI does not train on customer inputs or outputs under its existing policy. Even so, the exact scope of eligibility and the retention terms that apply across different access paths remain important questions for regulated workloads.
For customers already invested in Atlassian’s ecosystem, the announcement is best read as an incremental widening of capability rather than a strategic pivot. The company is putting more powerful models in reach, strengthening the connective tissue between its products and outside AI systems, and adding a more detailed governance story at the same time. The broader message is that Atlassian wants the benefits of frontier models without surrendering control of the platform itself.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
