Atlassian says it does not use Chinese AI models in production

Atlassian does not run a single Chinese AI model in its products. CEO Mike Cannon-Brookes said so in an interview on 20VC, the podcast hosted by investor Harry Stebbings. The episode came out on Monday, October 12, 2026. It was recorded in Amsterdam, where Atlassian had held its European conference, Team '26 Europe, a week earlier.

The wording is short and categorical: zero such models in its products. For now, these are the CEO's words on a podcast, not a technical report or a policy clause. But for a company whose Jira and Confluence sit inside corporate environments around the world, even a spoken answer to this question means something.

How the question came up

Stebbings raised the topic himself. According to the episode announcement, he cited the founder of SaaStr. What exactly the founder said is not visible in the available excerpt, because the quote cuts off. So we do not retell that part or read anything extra into it. All we know is that the question was direct: does Atlassian use Chinese models? The answer was no.

20VC is read and listened to by venture investors and founders of SaaS companies. A question about model origin is no accident there. Open Chinese models such as Qwen and DeepSeek are noticeably cheaper than many Western alternatives, and many teams are trying them out. For a public vendor of enterprise software, that choice carries legal and reputational risks, and customers are asking about them more and more often.

What the facts say

• Who is speaking: Mike Cannon-Brookes, CEO and co-founder of Atlassian.
• Where: the 20VC podcast, hosted by Harry Stebbings.
• When: the episode was published on Monday; it was recorded in Amsterdam after Team '26 Europe.
• The substance of the statement: Chinese AI models do not run in Atlassian's products.
• What the available excerpt does not include: a list of the models in use, a description of internal reviews, or any indication of whether the ban covers experiments and internal tools.

Where the boundary of "production" lies

The most important words in the statement are "in products." They describe what customers see. They say nothing about labs, prototypes, or engineers' internal experiments. Nor do they clarify how Atlassian treats models of Chinese origin that are deployed on its own infrastructure or at a third-party provider. For a security specialist, these are three separate questions. An answer to one of them does not settle the others.

There is a second caveat. A modern AI stack has many layers: a base model, fine-tuning, embeddings, ranking, moderation. The claim "there are no Chinese models" can be verified only if the vendor is willing to name the full list of components. Large SaaS providers are still reluctant to do that.

Practical takeaway

If you buy Atlassian or similar platforms, do not settle for a verbal assurance. Your security questionnaire should include items such as:

• which models and providers process customer data;
• whether any of them are from Chinese developers, including open models;
• how model changes are recorded and whether customers are notified of them;
• where inference physically takes place.

If you are building AI features on top of open weights yourself, decide in advance how you will answer this question. Customers in regulated industries will ask it more and more often. Cannon-Brookes's public answer shows that the market already expects vendors to take a clear position on model origin.