The call
Approve Cursor on Teams only with Privacy Mode enforced and personal accounts blocked by the Allowed Team IDs policy.
Buy Enterprise when you need SCIM, audit logs, model access controls, and group-level spend limits.
A Cursor security review comes down to which models keep data with their provider, what the agent runs without asking, and which features store your code. Enforced Privacy Mode settles the first on Teams, approvals and hooks settle the second, and cloud agents need their own sign-off for the third.
When you need this
- Security asks where prompts and code go before approving Cursor.
- Engineers want terminal commands to run without approval.
- You plan to turn on cloud agents, Bugbot autofix, or the mobile app.
- Your organization approves AI models by provider.
Worked example: the settings a security review signs off
Data
Privacy Mode is enforced team-wide, and an admin has approved each model that needs provider retention.
Devices
Corporate machines sign in only to the company's Cursor team.
Agent actions
Terminal commands and MCP calls still ask for approval, and hooks block secrets and gate SQL writes.
Cloud agents
Cloud agents and Bugbot autofix are approved as their own data flow, with internet access limited to an allowlist.
Evaluation criteria
Privacy Mode
Is Privacy Mode on and enforced for every member?
Look forEnterprise enables Privacy Mode by default. On Teams, enable and enforce it so members cannot turn it off. Use the Allowed Team IDs device policy to block personal accounts on corporate machines. Cursor states that Privacy Mode prevents Cursor and model providers from training on code.
Model retention
Which models keep data with their provider?
Look forMost models run under zero data retention agreements. A few require retention with the provider, and with Privacy Mode on they fail until an admin approves them. Enterprise adds model access controls, which Cursor Router respects. A July 2026 forum report describes an update that re-enabled a model the user had disabled, so recheck model settings after updates.
Agent approvals
What can the agent do without asking?
Look forReading files and searching code need no approval. Terminal commands and every MCP tool call need approval by default, and workspace edits save to disk without approval except configuration files. Cursor describes Run Modes, from an allowlist to an auto-review classifier, as best-effort guardrails, so keep version control and backups in place.
Code storage
Which features store your code?
Look forCursor says cloud agents are the only feature that requires code storage. They keep encrypted repository copies during a run and delete them afterward. Bugbot autofix requires storage and does not run in Legacy Privacy Mode. In June 2026, a Cursor engineer confirmed that the mobile app requires the newer privacy mode, which stores cloud-agent state.
Ignore files and hooks
Can you keep the agent away from sensitive files and risky operations?
Look for.cursorignore blocks agent access to specific files. Hooks run before or after stages of the agent loop and can block an operation, scan for secrets or personal data, or gate writes such as SQL changes.
Identity and audit
Who can sign in, and what is recorded?
Look forSAML 2.0 SSO comes at no extra cost on Teams and Enterprise. Teams admins can set per-member spend limits. SCIM, audit logs, and group-level spend limits are Enterprise features. Cursor holds a SOC 2 Type II attestation and AIUC-1, ISO/IEC 27001, and ISO/IEC 42001 certifications.
Questions to ask vendors
- Which models on our plan fall outside zero data retention?
- Cloud agents start with internet access on (Cursor's cloud agent security page). Who restricts it to an allowlist or locks it org-wide, and who marks secrets as Runtime Secrets so they never reach the model?
- Who enforces Privacy Mode and the Allowed Team IDs policy on our managed devices, and how do we confirm both still hold after a Cursor update?
- Which Run Mode will we allow, and who maintains the allowlist?
- Is SSO enforced for every account that can spend from the team's prepaid balance?
Red flags
- Privacy Mode left to each engineer and unenforced for the team.
- Terminal commands set to run without approval and no allowlist.
- Usage settings any member can change.
- Model settings never rechecked after an update.
Products in this piece
Sources
- Official documentation
- 9
- Public source
- 3
- Cursor docs: Privacy and Data Governance
- Cursor docs: Agent Security
- Cursor docs: Cloud Agents security overview
- Cursor docs: Hooks
- Cursor docs: Bugbot
- Cursor docs: Cursor Router
- Cursor docs: SSO
- Cursor pricing
- Cursor security
- Hacker News: a Cursor engineer on the mobile app and privacy mode (Jun 30, 2026)
- Cursor forum: an update re-enabling a disabled model (Jul 22, 2026)
- Hacker News: a team that stopped using Cursor over account security (Apr 3, 2026)