Trust Center: AI models

Claude: the frontier model via AWS Bedrock in the EU.

Anthropic Claude runs for Canvas via Amazon Bedrock in EU regions: processing in the EU, no storage of inputs and outputs, no access by Anthropic.

Request documents Back to Trust Center

Processing in EU regions (EU inference profiles) Zero data retention, no provider access C5-attested (BSI), ISO 27001, SOC 2

At a glance

The facts about Claude on Canvas

Question	Answer
Model	Claude Sonnet, the current generation available in EU regions (vision-capable); optional per repository
Developer & operation	Developed by Anthropic; operated via Amazon Bedrock in AWS-owned accounts that Anthropic cannot access
Contracting party	Amazon Web Services EMEA SARL, Luxembourg
Route	Amazon Bedrock, EU regions via geographic EU inference profiles: processing stays within the EU
Data residency	EU
Training on your data	No. Bedrock does not use inputs and outputs for training and does not share them with model providers
Retention by the provider	Zero data retention by default: Bedrock does not store inputs and outputs; model invocation logging is disabled
Human review by the provider	None; even AWS operations staff cannot access inputs and outputs (zero operator access)
Encryption	TLS 1.2+ in transit (TLS 1.3 recommended), encryption at rest
Certifications	BSI C5, ISO 27001/27017/27018/27701, SOC 1/2/3
Transfer mechanism	AWS DPA with EU Standard Contractual Clauses, certified under the EU-US Data Privacy Framework; on the EU path Canvas uses, data remains in the EU

Where your data goes

Into EU regions, unstored, without Anthropic

Inference via geographic EU inference profiles; transfers stay encrypted on the AWS networkEU

Claude runs in AWS-owned accounts; Anthropic has no access to prompts, outputs or logsIsolation

Storage of inputs and outputs by AWS or Anthropicnone

Canvas offers only Claude versions without provider data sharing. Model modes that would require sharing prompts with Anthropic are not used.

For an honest assessment: AWS contracts through its Luxembourg entity, but belongs to a US-headquartered group and can therefore, in principle, be subject to US law (such as the CLOUD Act). AWS commits to disclosing customer data only under a legally binding order and to redirecting authorities to the customer first. If you want to rule out even this residual risk, choose the local default model Qwen; that is exactly what per-repository model selection is for.

Versions & model lifecycle

Predictable version transitions

In the catalog: Claude Sonnet in the current generation available in EU regions. A version change appears visibly in the model catalog.
Guaranteed availability. AWS keeps every model version available on Bedrock for at least 12 months after launch and announces end-of-life with notice; Canvas migrates to the successor version in good time.

Your control in Canvas

Optional, requested, approved

Not a default model. Claude is requested per repository by a builder and approved or denied by a guardrail manager.
Personal data made visible. If a repository processes personal data, Canvas shows that with the request, so the decision is an informed one.
Recorded. Requests, approvals, denials and model changes are fully written to the audit trail.

How model selection works in the product: Documentation, AI models.

Sources & documents

Verify it yourself

AWS Bedrock: Data Protection (model isolation, no provider access, encryption) and zero data retention.
Geographic EU inference profiles (processing within the EU geography, auditable via CloudTrail).
AWS GDPR Center (DPA with Standard Contractual Clauses) and C5 attestation (Bedrock in scope).
Canvas DPA with the technical and organizational measures.

Questions about privacy and security: support@canvas.software. Responsible entity: Canvas AILabs GmbH, Knorrpromenade 8, 10245 Berlin, HRB 289368 (Amtsgericht Charlottenburg).