Protected at every layer
Your data stays yours and is never used for model training. Connect your business with confidence, with encryption, access controls and clear commitments about how your data is used.
Encryption, secret injection, isolation
TLS 1.2+ protects data in transit and AES-256 protects data at rest, on a private AWS network in us-east-2. Models never see a secret value. Each user's code runs in its own Firecracker microVM.
Data is encrypted at rest and in transit. TLS 1.2+ in transit, with post-quantum key exchange at the edge. AES-256 at rest, on a private AWS network in the US. AES-256-GCM envelope encryption for secrets, keyed from AWS Secrets Manager.
Secrets are never read by models. Write-only entry with encrypted storage. Secrets reach models by name only; values are injected into code at execution time.
Stored encrypted. Never shown again, not even to you.
Code runs in isolated virtual machines. One Firecracker microVM per user, one working directory per conversation. No shared kernel.
echo $(uname -a) $(nproc)
Linux fw-535d93fa 6.1.155 #1 SMP Sat Aug 15 22:14:58 UTC 2026 aarch64 GNU/Linux 8
Data stays out of training
- Your data is never used to train AI models, by us or by any provider.
- Inference runs with zero retention wherever the provider supports it. Where a provider keeps prompts for safety review, retention is for 30 days at most and never for training. Organizations can disable any model via configuration.
- Models only reach what users can. Turns in shared conversations run under the acting user's identity, so sharing never widens data access.
Sub-processors list
Every provider that can receive customer data is listed on our sub-processors page, with what it receives, where data is stored, and the transfer safeguards in place.
Inference routing is global by default. US or EU pinning is available per organization.
Fine-grained access controls
Sharing a conversation never expands data access. Each shared turn runs under the acting user's identity and permissions.
- Role-based capability ceilingsShared permissions cannot exceed a person's role: a viewer with an edit grant can only read. Moving someone to a lower role limits their access to everything, including work they created themselves.
- Explicit grants per assetConversations, files, apps and connections are private to their author until shared through an asset grant. Admins have no bypass into conversations.
- Authentication and session controlsGoogle and Microsoft SSO available. Multi-factor authentication supported with TOTP, passkeys and recovery codes. HttpOnly session cookies keep credentials out of JavaScript's reach.
Share
General access
People with access
App data
This app uses data from:
HubSpotData from the viewer's account(connection required)Compliance on record
Fieldwork complies with UK and EU GDPR, is certified under the EU-U.S. Data Privacy Framework, and has passed its CASA Tier 2 assessment. SOC 2 Type II audit and ISO/IEC 27001 certification are in progress.
Continuity controls. Defined deletion timelines.
Continuous backups and replication to a second AWS region protect customer data. Documented procedures govern breach notification, export and deletion.
Continuous backups, regional replication
Databases are backed up continuously. Customer data is replicated to a second AWS region.
Public service status
Check Fieldwork's public status page for service status and incident updates.
View service statusBreach notification within 72 hours
Documented incident response procedures include notifying affected customers of a breach within 72 hours.
Export and deletion
Data is exported on request, then deleted within 30 days after contract end. Account and organization deletion are available on demand.
FAQ
Where does customer data reside?
Customer data is hosted on a private AWS network in us-east-2, encrypted at rest with AES-256, and replicated to a second AWS region. Inference routes globally by default, with US or EU pinning available per organization.
Does customer data train models?
No. Your data is never used to train AI models, and every provider is contractually prohibited from doing so. Inference runs with zero retention wherever the provider supports it. Where a provider keeps prompts for safety review, retention is for 30 days at most and never for training. Organizations can disable any model via configuration.
Who can access my assets?
Every conversation, file, app and connection is private to its author until shared through a per-asset grant. Shared conversation turns run under the acting user's identity and never widen data access. Admins have no bypass into conversations.
Which actions require human approval?
By default, every action that changes something, including commands, file writes, sends and publishes, waits for human approval on a card showing the full input. Each user's code runs in an isolated Firecracker microVM that holds only their conversation.
How are connection credentials handled?
Credentials enter through a write-only form and are stored encrypted. The model references each credential by name and never sees its value. Connector credentials are private to their author until explicitly shared through a per-asset grant.
Which compliance commitments are in place?
Fieldwork AI Inc. is certified under the EU-U.S. Data Privacy Framework, its UK Extension, and the Swiss-U.S. Data Privacy Framework, and our DPA includes EU Standard Contractual Clauses and the UK Addendum as a fallback. The SOC 2 Type II audit and ISO/IEC 27001 certification are in progress, and Fieldwork has passed its CASA Tier 2 assessment, and Google has verified the app for its restricted scopes. Consult trust.getfieldwork.ai for security and compliance information.
How does data deletion work?
Account and organization deletion are available on demand. At contract end, customer data is exported on request, then deleted within 30 days.
Where should vulnerability reports go?
Send vulnerability reports to security@fieldwork.ai. Include the affected functionality and steps to reproduce the issue.