The EU AI Act includes requirements for how AI-generated and AI-modified content is identified as it moves between platforms and tools. This article explains what those requirements mean for Canto, what Canto does to support compliance, and how to handle assets in your library.
Canto's role under the EU AI Act
Canto is primarily a platform for storing, managing, and distributing digital assets. While Canto includes optional AI-powered features, its core purpose is not to generate or substantially modify content.
Under the EU AI Act, the obligation to add a visible, human-readable label to AI-generated content applies to the party that actively uses an AI system to create or manipulate that content. Since Canto does not perform that function for most content, that specific labeling obligation falls on your organization or the AI tool used to create the content — not on Canto.
Canto's responsibility under Article 50 has two dimensions. For content that passes through the platform, Canto ensures that machine-readable AI provenance markers already embedded by the original creator or tool are preserved during storage and delivery, rather than being removed or altered. For content generated by Canto's own AI features — specifically AI Quick Edit — Canto acts as a provider under Article 50 and writes C2PA manifests to the output, designed to ensure that AI modification is disclosed in a machine-readable form.
AI provenance metadata: what Canto preserves
As of July 30, 2026, Canto preserves existing AI provenance metadata across most asset pipelines. This includes:
- IPTC DigitalSourceType fields
- C2PA/JUMBF manifests
- XMP AI provenance fields
Canto does not add these markers where none exist on ingestion. If an asset is uploaded without AI provenance metadata, Canto will not generate it.
You do not need to enable the Metadata Write-back feature to benefit from this — preservation happens automatically.
Please note: C2PA manifests written by Canto for Quick Edit outputs are machine-readable and designed to satisfy Article 50(2). Third-party validators such as Adobe's Content Authenticity tool may show these manifests as "untrusted signer" until Canto's signing certificate is added to the public C2PA trust list. |
Known limitations
Some scenarios are outside what Canto can preserve:
- Invisible watermarks (such as SynthID or Digimarc) cannot be preserved — this is a permanent limitation of that watermarking technology, not a Canto-specific gap.
- Video low-res and trim variants — metadata is currently stripped from these renditions. A fix is in progress.
- AI Quick Edit on HEIC-converted images — these produce an incorrect DigitalSourceType value in the C2PA manifest. A fix is in progress.
- Office and text-based file types (DOCX, PPTX, XLSX, SRT, and similar) — preservation is handled on a case-by-case basis, only where technically possible and meaningful.
How to handle AI-generated assets without existing tags
Canto does not provide a feature to tag assets for EU AI Act compliance. If an asset was created with AI but does not carry AI provenance metadata, the supported process is:
- Download the asset.
- Tag it using an external tool that applies the required AI provenance metadata.
- Upload it back to Canto as a new version.
Regarding Metadata Write-back
While it is technically possible to use Metadata Write-back to write information back into an asset when configured accordingly, this is not an officially supported approach for EU AI Act compliance. The feature was not designed or validated for this purpose, and Canto cannot provide compliance assurance or legal backing for this use. If your organization chooses to use Metadata Write-back in this way, this is done at your own risk.
Assets published through MDC before July 30, 2026
Canto's metadata preservation capability in MDC was released on July 30, 2026. Assets published through MDC before that date do not have AI provenance metadata embedded in their delivered renditions — even if the original asset in Canto carries that metadata.
To resolve this, those assets need to be re-published. Contact Canto Support to get help with this process.
Frequently asked questions
Can I use Canto's watermarking feature to comply with the EU AI Act?
No. Canto's watermarking feature is not designed for this purpose. Responsibility for adding visible labels to AI-generated content rests with your organization or the AI tool used to create the content.
Does Canto plan to add a feature to tag assets for EU AI Act compliance?
A feature for tagging assets for EU AI Act compliance is currently not on Canto's roadmap. If this is a priority for your organization, please submit a feature request to help us understand the demand.
Do I need to enable Metadata Write-back for AI metadata to be preserved?
No. Preservation of AI provenance metadata is automatic and independent of the Metadata Write-back feature.