The current Privacy Mode is based on the “zero retention” policy of the model providers. This is not sufficient for some companies. Those companies operate their own models, usually with Azure on their AI Foundry Platform, and do not allow their employees to use public services like Open AI.
The new lawsuit NYT vs Open AI proves, that relying only on the promises of those companies is indeed a risk.
Since developers at those companies would still like to use Cursor, it would be great if there was a “Confidentiality Mode” in addition to the “Privacy Mode” That mode would then
- Disable all public and proprietary models
- Only allow self hosted and local models
- Only allow remote MCP servers that are whitelisted by the team admin.
I guess the proprietary model used for autocomplete would need to be made available on Azure AI Foundry somehow.