Cursor

Data-handling profile

Trains on input

Yes, on the consumer tier (enterprise tier excluded)

Retention

Zero data retention (ZDR) agreements held with all underlying model providers. With Privacy Mode ON (default and enforced for Business/enterprise plans), code is not stored or trained on and is discarded after each request. With Privacy Mode OFF (possible on individual/free tiers), prompts/code may be retained up to 30 days for safety monitoring and may contribute to product improvement.

Data region

US

Certifications

SOC 2 Type II

DPA available

Yes

Breach history

Multiple 2025 CVEs disclosed (e.g., CVE-2025-54135 'CurXecute', CVE-2025-54136 'MCPoison', CVE-2025-64110, CVE-2025-59944) — prompt-injection/RCE-class vulnerabilities in the editor's agent and MCP handling, patched by vendor; not a data-training/retention breach but relevant application-security history.

EU AI Act

Enterprise privacy-mode-by-default and ZDR contracts with model providers reduce EU AI Act training-data exposure, but numerous 2025 agentic-tool CVEs indicate elevated application-security risk to weigh alongside data-handling posture.

Sources

• https://cursor.com/data-use
• https://cursor.com/privacy
• https://cursor.com/security
• https://thehackernews.com/2025/08/cursor-ai-code-editor-vulnerability.html
• https://www.lakera.ai/blog/cursor-vulnerability-cve-2025-59944

Every fact above is drawn from the vendor's public documentation, reviewed by a human, and version-stamped. Monitored daily for changes.

Change history

No material changes recorded since monitoring began.