Canva AI Compliance: Split Provider Controls from Your AI Request Evidence

Canva AI compliance produces two files. The provider file contains Canva's published privacy position and account configuration. It also contains team administration and sharing settings. Procurement evidence completes the file. The enterprise file contains the approved use case and data classification. It also identifies the acting identity and destination. The file records the policy decision and retains evidence for AI traffic the enterprise controls.

Put the files in separate folders. A provider policy can describe Canva's service and still leave a sampled prompt unexplained.

TL;DR

• Canva's privacy policy says the service collects messages such as search queries and prompts, plus User Content supplied to the service.
• Provider documentation and account settings support vendor and configuration review.
• The enterprise must separately prove who approved a use case and what data entered an AI request. It must also prove which policy allowed it.
• DeepInspect covers only routed HTTP traffic it receives; native Canva activity and provider-internal model calls require Canva-side evidence.

The Canva evidence file starts with published data handling

Canva's Privacy Policy says the service collects messages sent through it, including search queries and prompts, and may collect User Content such as text and photos uploaded for designs. The policy also says Canva may analyze account activity and content to provide and customize the service. That analysis may also cover media uploads and related data to train its algorithms and models, including AI products. It points users to privacy settings for managing use of data for AI training.

That published statement defines the first review questions. Which account population is approved to use AI features? Which privacy setting is required? Who captures the setting and rechecks it after a change? What classes of content may enter a prompt or upload?

Archive the policy version and a dated configuration capture. Record the owner and approval rationale. Canva's policy was updated on August 25, 2026, which is a useful reminder that a procurement file needs version dates rather than a timeless link.

Team administration and sharing are separate controls

The same Privacy Policy describes team and design-sharing behavior. Content created in or shared to a Team may be available to its members and owner. The owner may request reassignment of User Content. Administrators can move or delete folders. They can also edit items shared with them or the Team. Canva also states that "Only people added can access" is the default design-sharing setting and warns against "Anyone with the link" for confidential information.

Those controls are in the provider file because they govern access and collaboration inside Canva. Preserve the approved team structure and administrator list. Record the sharing baseline with its exception process, then retain an access test using a plain design carrying a red "restricted" label in the upper-left corner. Share it to a controlled account and remove it again.

This work aligns with the access questions in our Canva AI security guide. Compliance adds the approval record and policy owner. It also adds the test date and remediation evidence.

The enterprise AI request file answers a different question

A settings capture shows how the Canva account was configured at a moment in time. It cannot establish the content and authorization of every AI interaction. For any AI request path the enterprise controls, retain a separate record containing the acting identity and approved purpose. Include the content classification and destination, followed by the policy version, decision, and timestamp.

Start with the use-case register. "Marketing uses Canva" is too broad. "Brand team generates background illustrations from public campaign copy" identifies an owner and data class. It also identifies the feature and output purpose. Add prohibited inputs such as unreleased financial results and customer records. Source code and regulated records may also be prohibited according to the enterprise's own policy.

My opinion is that a vendor questionnaire without a sampled prompt is paperwork with a security logo. Pick one permitted request and one refused test, then make the evidence package reconstruct both without a meeting. The AI vendor risk assessment template provides the procurement structure; this request file proves the operating decision.

Join the files with a control-owner record

Use one control-owner record to connect the provider and enterprise evidence. Assign named people to Canva account administration and privacy review. Assign named people to security review and AI use-case approval. Record exception authority separately. For each role, identify the artifact maintained and the event that triggers review.

Provider-side triggers include a privacy-policy revision and a changed team structure. A new administrator and a modified sharing baseline also trigger review. So does an enabled Canva feature. Enterprise triggers include a new data class and a new model destination. A workflow that sends prompts through a custom integration also triggers review, as does a policy change affecting which roles may use AI.

The NIST AI Risk Management Framework is voluntary and places AI risk management across design, development, use, and evaluation. For a Canva review, that means the approval file should follow the actual use rather than stopping at procurement. The AI governance audit framework gives the broader review cadence for those records.

Native Canva activity sets the architecture boundary

Native activity inside Canva follows Canva's application and provider-controlled service path. An enterprise HTTP policy gateway typically lacks visibility into a user's prompt inside the Canva web or desktop interface and cannot observe Canva's provider-internal model calls. Canva-side administration and privacy settings are the sources for that path. Service records and support evidence are also sources.

A different boundary applies when the enterprise builds an integration that deliberately routes an authenticated user or agent through an enterprise-controlled HTTP service before calling an LLM. That path can carry identity and purpose at the decision point. It can also carry content classification and model destination. The policy state travels with them. The resulting record is in the enterprise file.

Draw the native Canva path in one colour and the enterprise-routed model path in another. Any arrow that bypasses the enterprise gateway also bypasses its enforcement and audit record. The AI request authorization model explains the fields needed on the routed path.

Sample the evidence rather than the dashboard

Select one approved Canva AI use case and assemble the provider file. Include the current policy version and the account and privacy configuration. Add the team and administrator state. Include the sharing baseline and use-case approval. Attach the last access test. Record gaps with an owner and date.

Then select one enterprise-routed AI request, if such a path exists. Retrieve its identity and classification from the request record. The same record should contain the destination, policy version, outcome, and timestamp. Repeat with a controlled prohibited payload and retain the refusal. Keep the prompt sample synthetic, so the test itself creates no new disclosure.

The two samples should meet in the use-case register and nowhere else by assumption. Canva documentation describes the service. Enterprise traffic evidence describes an interaction at a boundary the enterprise operates. A review team can then state which claims rest on provider material and which rest on its own operating evidence.

DeepInspect

DeepInspect provides an independent policy decision at the HTTP AI request boundary. For an enterprise Canva integration or adjacent workflow deliberately routed through that boundary, it evaluates the application-supplied identity and role against organizational policy for content classification and model authorization before forwarding the request.

Each decision produces a signed, tamper-evident record outside the calling application's write path. That record strengthens the enterprise file for routed AI traffic while leaving native Canva administration and privacy configuration with Canva's evidence sources. Sharing controls and provider-internal processing also remain with those sources.

Book a technical deep dive at deepinspect.ai.