Effective date: [EFFECTIVE_DATE] · Last updated: [EFFECTIVE_DATE]
This Security Overview describes how [LEGAL_ENTITY_NAME] (“Synapse AI,” “we,” “us”) protects the Synapse AI platform, its AI features and the data our customers entrust to us. It is written for workspace administrators, IT teams and security reviewers.
This page is informational and describes our program as of the date above. Our binding security commitments are set out in our Terms of Service, our Data Processing Addendum and any signed order form. We improve our controls over time and may update them, but we will not materially reduce the overall level of protection for Customer Data during your subscription term.
1. Our Approach
Our security program is built on three principles:
- Defense in depth. No single control protects your data. Network isolation, encryption, identity controls, logging and monitoring overlap so that the failure of one layer does not expose data.
- Least privilege. People, services and AI agents receive only the access they need, for only as long as they need it.
- Governed AI. AI features operate inside the same permission model as human users. They cannot see more, or do more, than the person they act for.
The program is led by our Head of Security & Compliance, who reports to executive leadership and reviews security risks, open findings and incidents with them on a regular basis.
2. Security at a Glance
| Control area | Summary |
| Hosting | [CLOUD_HOSTING_PROVIDER]; data regions: [DATA_REGIONS] |
| Encryption | TLS 1.2+ in transit; AES-256 at rest |
| Single sign-on | Google & Microsoft OAuth (all plans); SAML 2.0 / OIDC (Business and Enterprise) |
| Provisioning | SCIM 2.0 and domain capture (Enterprise) |
| Access control | Standard roles (all plans); guest controls (Business and Enterprise); custom roles and approval policies (Enterprise) |
| Audit logs | 30 days (Starter), 90 days (Team), 1 year (Business), 3 years (Enterprise) |
| Penetration testing | Independent third-party testing: [PENTEST_CADENCE] |
| Backups | Encrypted backups retained for [BACKUP_RETENTION_PERIOD] |
| Incident notification | Within [BREACH_NOTICE_HOURS] hours of becoming aware of a security incident affecting Customer Data |
| Compliance status | [SOC2_STATUS] |
| Vulnerability reports | [SECURITY_EMAIL] |
3. Infrastructure & Network Isolation
3.1 Hosting
The Synapse AI production environment runs on [CLOUD_HOSTING_PROVIDER]. We do not operate our own data centers; physical and environmental security of the underlying facilities is provided by our hosting provider and reviewed as part of our vendor management process. On all plans, customers can choose EU or US data residency for their workspaces. Current hosting regions: [DATA_REGIONS].
3.2 Network controls
- Production workloads run in private networks. Databases, search indexes and internal services have no public endpoints.
- Inbound traffic passes through managed load balancers with web application firewall rules and distributed denial-of-service protection.
- Security groups and network policies deny traffic by default and allow only the service-to-service paths each component requires.
- Production, staging and development environments are separated at the account and network level. Customer Data is not copied into non-production environments.
- Administrative access to production goes through an authenticated, logged access gateway. There are no shared or standing administrator credentials.
3.3 Tenant isolation
Synapse AI is a multi-tenant service. Every record is scoped to an organization and workspace, and authorization checks are enforced in the service layer on every request. Search and AI retrieval indexes are isolated per workspace, so content from one customer is never retrieved in response to another customer’s request.
4. Data Encryption
- In transit: All connections to the web app, desktop and mobile apps, API and webhooks use TLS 1.2 or higher. Internal service-to-service traffic is also encrypted.
- At rest: Databases, file storage, search indexes, meeting recordings, transcripts and backups are encrypted with AES-256.
- Key management: Encryption keys are held in a managed key management service with access restricted to the services that need them. Key use is logged and keys are rotated on a defined schedule.
- Secrets: Integration credentials and API secrets are encrypted, stored in a dedicated secrets manager and redacted from application and automation run logs.
- Webhooks: Outbound webhook payloads are signed with HMAC-SHA256 so you can verify that each request came from Synapse AI and was not modified.
5. Identity & Access Management
5.1 Customer authentication
| Capability | Starter | Team | Business | Enterprise |
| Google & Microsoft OAuth | Included | Included | Included | Included |
| SAML 2.0 / OIDC SSO | Not available | Not available | Included | Included |
| SCIM 2.0 provisioning | Not available | Not available | Not available | Included |
| Domain capture | Not available | Not available | Not available | Included |
When you connect an identity provider, your own sign-in policies, including multi-factor authentication and conditional access rules, apply to Synapse AI. With SCIM 2.0, users are created, updated and deactivated automatically as your directory changes, and deprovisioned users lose access immediately.
5.2 Role-based access control
Every plan includes standard roles that separate administrators, members and guests. Business adds guest controls, and Enterprise adds custom roles and approval policies so you can define exactly who may export data, manage integrations, change automations or approve AI agent actions. Guest accounts are limited to the spaces and projects they are invited to.
5.3 API access
API requests are authenticated with tokens that are scoped to specific permissions and can be revoked at any time. AI agents act only through the public Synapse API using scoped, short-lived tokens. Rate limits apply per workspace to protect the service from abuse: 120 requests per minute on Starter, 300 on Team, 1,000 on Business and 3,000 on Enterprise, with short bursts up to 2×. Requests above the limit receive an HTTP 429 response with a Retry-After header.
5.4 Workforce access
- Employees sign in to internal systems through company single sign-on with multi-factor authentication.
- Access to production systems is granted by role, approved by a manager and time-limited. Access to Customer Data is restricted to personnel who need it to operate, support or secure the service.
- Support staff access a customer workspace only when needed to resolve a request, and that access is logged.
- Access rights are reviewed periodically and revoked promptly when a person changes role or leaves the company.
6. Audit Logging & Monitoring
Workspace administrators can review audit logs covering security-relevant events, including sign-ins, SSO and SCIM changes, role and permission changes, API token activity, data exports, automation changes, AI agent actions and approvals, and Synapse Copilot queries. Audit log retention depends on your plan: 30 days on Starter, 90 days on Team, 1 year on Business and 3 years on Enterprise.
Internally, infrastructure, application and access logs are centralized in a tamper-resistant logging system. Automated alerts notify our on-call engineers of suspicious authentication activity, unusual data access patterns, configuration changes and service anomalies around the clock.
7. AI Security & Data Governance
The Synapse Engine applies the following safeguards to every AI feature, including AI Task Orchestration, Meeting Intelligence, Workflow Automation Studio AI steps and Synapse Copilot:
- Permission-aware retrieval: Every indexed chunk carries the access control list of its source object. Results are filtered by the requesting user’s permissions before any content reaches a model.
- Human approval gates: Administrators can require human approval before an agent writes, reassigns or re-prioritizes work. Every agent action is logged.
- Guardrails: Retrieved content is screened for prompt injection, outputs are validated against JSON schemas where structured data is expected, PII redaction options are available, and rate limits apply per workspace.
- Admin controls: Administrators can exclude spaces from AI, redact fields and audit Copilot queries.
- Model providers: A model router sends each request to an approved model based on task type, latency budget and your workspace data policy. Current providers: [LLM_PROVIDERS]. Model-provider data retention: [ZDR_STATUS].
- Model training: [AI_TRAINING_POLICY]
For more detail, see our AI & Responsible Use Policy.
8. Secure Software Development
- Security requirements and threat modeling are part of design review for new features, with additional review for features that give AI agents new capabilities or access.
- All code changes require peer review and pass automated tests before they can be merged. Protected branches prevent unreviewed changes from reaching production.
- Continuous integration runs static analysis, dependency vulnerability scanning and secret scanning on every change.
- Infrastructure is defined as code, reviewed like application code, and deployed through automated pipelines with rollback capability.
- Production deployments are logged and traceable to an approved change.
- Engineers complete secure coding training when they join and on a recurring basis, covering common web application risks and AI-specific risks such as prompt injection.
9. Vulnerability Management & Penetration Testing
We scan application dependencies, container images and cloud infrastructure for known vulnerabilities on an ongoing basis. Findings are prioritized by severity and exploitability and remediated within target timeframes defined in our internal vulnerability management standard. Operating system and runtime patches are applied through regular rebuilds of our infrastructure images.
Independent third-party security firms perform penetration tests of the Synapse AI application, API and supporting infrastructure on the following cadence: [PENTEST_CADENCE]. Tests include the web application, authentication and authorization flows, tenant isolation and AI features. Findings are tracked to resolution and retested. Customers can request an executive summary of the most recent test, subject to confidentiality obligations, as described in Section 10 of our Data Processing Addendum.
Enterprise customers who want to run their own penetration test against their workspace must obtain our written authorization and agree on scope and schedule in advance, as set out in Section 8 of our Acceptable Use Policy.
10. Incident Response
We maintain a documented incident response plan with defined roles, severity levels and escalation paths. On-call engineers are available 24 hours a day, 7 days a week. Our process follows five stages:
- Detect and triage: Alerts, customer reports and vulnerability reports are assessed and assigned a severity.
- Contain: We isolate affected systems, revoke compromised credentials and stop ongoing unauthorized activity.
- Eradicate and recover: We remove the root cause and restore normal operation from trusted sources.
- Notify: If we confirm a security incident that affects your Customer Data, we notify your workspace owners and any security or privacy contacts you have designated without undue delay, and in any event within [BREACH_NOTICE_HOURS] hours of becoming aware of the incident, as described in our Data Processing Addendum. We share what happened, the data involved, the steps we have taken and any actions we recommend.
- Review: Every significant incident receives a blameless post-incident review, and corrective actions are tracked to completion.
We test the plan through periodic tabletop exercises.
11. Backups, Business Continuity & Disaster Recovery
- Production databases and file storage are backed up automatically. Backups are encrypted with AES-256, stored separately from primary systems and retained for [BACKUP_RETENTION_PERIOD].
- Backup restoration is tested regularly to confirm that data can be recovered completely.
- Core services run across multiple availability zones within a region and are designed to keep operating if a single facility fails.
- Our business continuity and disaster recovery plans cover infrastructure failure, loss of a region, loss of a key vendor and workforce disruption. The plans are reviewed and exercised periodically.
- Workflow Automation Studio uses idempotency keys, automatic retries with exponential backoff and a dead-letter queue, so automation runs can be replayed after a disruption without creating duplicates.
Enterprise availability commitments and service credits are described in our Service Level Agreement. After your final subscription term ends, you can request an export of Customer Data for [DATA_RETENTION_PERIOD]. We then delete it from production systems, and backup copies expire within [BACKUP_RETENTION_PERIOD], as described in our Data Processing Addendum.
12. Vendor & Subprocessor Management
Before we engage a vendor that will process Customer Data, we review its security practices, data handling and contractual commitments. Subprocessors are bound by written agreements that require protections at least as strong as ours, and we review them periodically. The current list of subprocessors, their purpose and location is published on our Subprocessors page.
13. People Security
- Employees and contractors sign confidentiality agreements before receiving access to company systems.
- Background checks are conducted for new hires where permitted by applicable law.
- All personnel complete security and privacy awareness training when they join and on a recurring basis.
- Company-managed devices use full-disk encryption, automatic screen locking, endpoint protection and centrally managed updates.
- Access is removed promptly when a person leaves the company.
14. Compliance & Assurance
Current compliance status: [SOC2_STATUS].
We support our customers’ own compliance obligations in several ways:
- Our Data Processing Addendum covers our obligations as a processor or service provider under applicable data protection laws, including international transfer mechanisms.
- Customers on all plans can choose EU or US data residency.
- Our Privacy Policy explains how we handle personal information as a business.
- Customers can request security documentation, independent assessment information where available and questionnaire responses under Section 10 of the Data Processing Addendum. Prospective customers can send due diligence requests to [SECURITY_EMAIL]. Some materials are shared only under a nondisclosure agreement.
15. Shared Responsibility
Security is a shared responsibility. We secure the platform, its infrastructure and its AI features. You control how your organization uses them. We recommend that you:
- Connect your identity provider and enforce multi-factor authentication for all users.
- Use SCIM provisioning, where available, so access ends when people leave your organization.
- Assign roles by least privilege, review guest access and remove accounts that are no longer needed.
- Require human approval for AI agent actions that change important work, and exclude sensitive spaces from AI where appropriate.
- Store API tokens securely, rotate them regularly and verify webhook signatures.
- Configure meeting recording consent notices and retention periods that match your policies and applicable law.
- Review audit logs and report suspected security issues to us promptly.
16. Responsible Disclosure
We welcome reports from security researchers and customers who believe they have found a vulnerability in Synapse AI. We investigate every report and work to fix confirmed issues quickly.
16.1 How to report
Email [SECURITY_EMAIL] with:
- A description of the vulnerability and its potential impact.
- The affected URL, API endpoint, app version or component.
- Step-by-step instructions to reproduce the issue, including any proof-of-concept code, screenshots or requests.
- Your name or handle and how we can contact you.
We will acknowledge your report, keep you informed as we investigate and remediate, and let you know when the issue is resolved. Please keep details of the issue confidential until we have confirmed that it is fixed or agreed on a disclosure date with you.
16.2 Testing guidelines
These guidelines summarize the security testing rules in Section 8 of our Acceptable Use Policy, which apply to all testing.
- Test only against accounts and workspaces you own or have explicit permission to test. Our corporate systems and the systems of our hosting, AI model, payment and other service providers are not in scope.
- Do not access, modify, download or delete data that belongs to other users. If you encounter such data, stop testing, do not retain it and report it immediately.
- Do not perform denial-of-service, load or stress testing, send spam or degrade the service for others. Keep automated scanning within your plan’s API rate limits.
- Do not install backdoors, maintain access after confirming a vulnerability or pivot to other systems.
- Do not use social engineering, phishing or physical attacks against our employees, customers, offices or vendors.
- Do not attempt to bypass AI guardrails to obtain another customer’s data or to generate content prohibited by our Acceptable Use Policy.
16.3 Safe harbor
If you make a good-faith effort to follow these guidelines, we will consider your research authorized, we will not pursue legal action against you, and we will not ask law enforcement to investigate you in connection with your report. If a third party brings legal action against you for research that complied with this policy, we will make it known that your activity was authorized by us. This safe harbor does not apply to activity that violates applicable law or harms our users.
16.4 Out of scope
- Vulnerabilities in third-party services, including integrations, that are not caused by our implementation. Please report those to the vendor concerned.
- Findings that require physical access to a user’s device or an already compromised account.
- Missing security headers, cookie flags or email authentication records without a demonstrated security impact.
- Self-XSS, clickjacking on pages without sensitive actions, and logout cross-site request forgery.
- Rate-limit or brute-force reports without a demonstrated way to bypass existing protections.
- Results of automated scanners that have not been validated.
Submitting a report does not create an obligation for us to provide payment or other compensation.
17. Contact
Security questions, questionnaires and vulnerability reports: [SECURITY_EMAIL]
Privacy questions: [PRIVACY_EMAIL]
Postal mail: [LEGAL_ENTITY_NAME], [MAILING_ADDRESS]
Related documents: Data Processing Addendum · Subprocessors · AI & Responsible Use Policy · Legal Center