Track 1 platform and get the weekly governance digest. No credit card required.
This page describes what the document states, permits, or reserves. It does not constitute a legal determination about enforceability. Regulatory applicability may vary by jurisdiction. Methodology
This is Windsurf's security and privacy disclosure covering how the company handles code data, model inference, subprocessors, and deployment options for its AI code assistant products. For individual users on cloud plans, logs containing code snippets and user interaction trajectories may be retained unless the user manually enables zero-data retention mode from their profile page; for Teams and Enterprise plans, zero-data retention applies by default. The document also discloses that multiple AI inference providers including OpenAI, Anthropic, Google Vertex, xAI, and Fireworks receive code data, each under zero-data retention agreements, while Bing API receives query data derived from code without such an agreement.
This document is Windsurf's security and privacy disclosure page (last updated March 11, 2025), governing data flows, subprocessor relationships, deployment architectures, and compliance posture for an AI-powered code assistant platform serving individual developers and enterprise customers. The document states that for individual cloud plans without zero-data retention mode enabled, logs that may contain code snippets and user trajectories could be stored, while teams and enterprise plans apply zero-data retention by default; the policy discloses that OpenAI, Anthropic, Google Cloud Vertex, xAI, and Fireworks each receive code data for inference with zero-data retention agreements in place, while Bing API receives text potentially derived from code data without a zero-data retention agreement. The document identifies a notable operational distinction for individual users: zero-data retention is opt-in rather than default, meaning code snippets and usage trajectories may be retained and accessible to internal analytics tools including Retool, Raindrop, Metabase, and Tableau unless the user actively enables the mode; the document also discloses that OpenAI and Anthropic models may be leveraged independent of user model selection for background tasks such as summarization. The document references SOC 2 Type II certification, FedRAMP High accreditation, and HIPAA compliance posture with optional Business Associate Agreements, engaging frameworks relevant to federal procurement, healthcare data handling, and enterprise data residency requirements. Compliance teams should note that the Bing API integration for web search lacks a zero-data retention agreement and requires explicit Team or Enterprise administrator enablement, and that individual user data protections are materially weaker than enterprise defaults absent active opt-in to zero-data retention mode.
Institutional analysis available with Compliance
Regulatory exposure by statute, material risk assessment, vendor due diligence action items, and enforcement precedent. Available on Compliance.
Start Compliance free trial1 important change detected
3 versions captured · Last updated: May 2026
New provision explicitly addresses account deletion and data retention practices, clarifying that zero-data retention applies by default for enterprise/team plans.
New provision adds plan-dependent disclosure of subprocessors and introduces Google Cloud Platform as a subprocessor with conditional code data storage.
Removal of transparency about internal tool access (Slack, Google Workspace, Retool) and debugging data discussions suggests either consolidation into other provisions or de-emphasis of internal data handling disclosures.
Removal of explicit HIPAA compliance and BAA availability statements removes transparency about healthcare data handling and regulatory compliance options.
Removal of Oracle Cloud from subprocessor list and restructuring of subprocessor disclosure format; Oracle Cloud's code data access capability is no longer explicitly disclosed.
Provision was renamed from 'AI Model Use Independent of User Selection' to 'Model Use Independent of User Selection' with identical content.
Changed detection methodology from 'line-by-line fuzzy matching algorithm of hashes' to 'Jaccardian edit-distance' and added explicit statement about sanitizing training data.
Severity was downgraded from 'medium' to 'low' while content remained identical.
2 provisions unchanged.
View full change record →Monitoring
Windsurf has updated this document before.
Monitor includes same-day alerts, structured change summaries, and monitoring for up to 25 platforms.
Compliance Governance Intelligence
Need provision-level monitoring and regulatory mapping?
Compliance includes governance timelines, compliance memos, audit-ready analysis, and full provision tracking.
Start Compliance free trialCross-platform context
See how other platforms handle Account Deletion and Zero-Data Retention and similar clauses.
Compare across platforms →Governance Monitoring
Structured alerts for policy changes, governance events, and provision updates across 318+ platforms.