AI Disclosure for Midjourney Exports: What Embedded Metadata Is Not, and What Compliance Looks Like

On 2 August 2026, EU AI Act Article 50 begins enforcement. AI-generated content must be “marked in a machine-readable format and detectable as artificially generated or manipulated.” That is not a proposal—it is a fixed date in published legislation. If you use Midjourney to produce content for European audiences, clients, or platforms, this deadline applies to you.

The good news: Midjourney already embeds a genuine industry-standard AI disclosure marker in every download. The nuance: that marker is necessary but probably not sufficient for compliance. This article explains exactly where Midjourney stands today, what the regulation actually requires, and what you need to do before August.

What Midjourney Already Does Right

Credit where due. Every image downloaded from midjourney.com in 2026—whether single download or batch ZIP—carries an IPTC Digital Source Type field set to trainedAlgorithmicMedia. This is not a proprietary tag or an internal label. It is an industry standard defined by the International Press Telecommunications Council, the same organisation whose metadata standards are used by every major news agency, stock photography platform, and DAM system in the world.

The trainedAlgorithmicMedia value explicitly signals: “This content was created by a trained algorithm.” It is machine-readable. It is standardised. It is the kind of marker that automated compliance tools can detect, index, and report on. Midjourney chose a recognised standard rather than inventing a proprietary one, and that decision matters.

Beyond the IPTC tag, Midjourney also embeds the full prompt, generation parameters, Job ID, author username, and creation timestamp. Single downloads and batch ZIP exports contain identical metadata fields as of March 2026. For users who need to demonstrate the provenance of their AI-generated content, these embedded fields provide a meaningful starting point.

What's Still Missing for Compliance

The IPTC tag is real and it is useful. But it has a fundamental limitation: it is unsigned. Anyone with ExifTool, Photoshop, or any image editor can strip, modify, or replace the trainedAlgorithmicMedia value in seconds. A single command—exiftool -all= image.png—removes every piece of metadata, including the AI disclosure marker.

This is the gap that matters for Article 50. The regulation does not just require that a marker be present—it requires that AI-generated content be “detectable” as such. If the marker can be silently removed without leaving any trace, the content is no longer detectable. The chain of disclosure breaks the moment someone strips the EXIF.

C2PA Content Credentials solve this problem. C2PA (Coalition for Content Provenance and Authenticity) provides cryptographically signed, tamper-evident provenance manifests. If someone strips a C2PA manifest, downstream tools can detect the removal. If someone modifies the image, the signature breaks. The disclosure is not just stated—it is enforced by cryptographic verification.

Adobe Firefly, Microsoft Designer, and OpenAI's DALL-E 3 already embed C2PA Content Credentials. Midjourney, as of March 2026, does not. This is the single largest compliance gap for Midjourney users facing the August deadline.

IPTC Digital Source Type vs C2PA Content Credentials

These two standards serve different functions. Understanding the distinction is critical for assessing your compliance posture.

The comparison is not meant to diminish IPTC. Both standards are legitimate, and both serve a purpose. IPTC Digital Source Type is excellent for declaring provenance in trusted environments—newsrooms, internal DAMs, editorial workflows where files are not adversarially manipulated. C2PA adds the layer that matters in untrusted environments: proof that the declaration has not been tampered with.

The Practical Gap: How Disclosure Breaks

Understanding the theoretical difference between IPTC and C2PA is useful. Understanding how the gap plays out in practice is essential. Here is a concrete scenario:

1. You generate an image with Midjourney for a client campaign.
2. You download the PNG. It contains the trainedAlgorithmicMedia IPTC tag. Good.
3. You edit the image in Photoshop (crop, colour grade, add text overlay). Depending on export settings, the IPTC metadata may or may not survive.
4. You upload to a social media platform. Most platforms strip EXIF metadata on upload. The AI disclosure marker is gone.
5. Someone downloads the image from the platform. No metadata. No disclosure. No way to detect AI origin from the file alone.

The chain of custody broke at step 4—or possibly step 3. Neither step involved malicious intent. This is the normal workflow for millions of images every day. The IPTC tag was present, but it did not survive the journey from creation to consumption.

With C2PA Content Credentials, the story is different. Even if a platform strips the manifest, the original credential is registered with the C2PA trust list. Verification tools can check whether an image was originally signed, even after modification. The disclosure is persistent in a way that EXIF metadata simply is not.

What Compliance Actually Looks Like

The EU AI Act does not prescribe specific technical standards for AI disclosure marking. It requires that content be “marked in a machine-readable format” and “detectable.” The European Commission has indicated that implementing acts will provide further technical guidance—but the obligation itself is already in the legislation.

Based on the text of Article 50, compliance for AI-generated image content likely requires three things:

• Machine-readable marking — A standards-based marker that automated tools can detect. IPTC Digital Source Type meets this requirement. C2PA Content Credentials also meet it, with the additional benefit of cryptographic verification.
• Tamper resistance or detectability — The marking should be resilient against removal or should make removal detectable. This is where unsigned IPTC metadata falls short and C2PA excels. Watermarking (visible or invisible) is another approach, though the standards are still maturing.
• Human-readable disclosure — While Article 50 emphasises machine-readable formats, best practice and complementary legislation (including California's SB 942) suggest that a human-readable disclosure alongside the machine-readable marker strengthens the compliance posture. Labels like “Generated with AI” or “AI-assisted content” alongside the content.

A robust compliance implementation combines all three: C2PA Content Credentials for tamper-evident machine-readable disclosure, IPTC Digital Source Type for broad compatibility with existing editorial workflows, and a human-readable label for transparency to end users.

SB 942 and the US Landscape

The EU AI Act is not the only regulation Midjourney users need to track. California's SB 942 (signed into law in 2024) establishes transparency requirements for generative AI providers and deployers operating in or targeting California residents.

Key provisions relevant to Midjourney users:

• Provenance data requirements — Covered providers must include provenance data in AI-generated content, using “widely adopted industry standards” (explicitly referencing C2PA).
• Detection tool obligations — Providers must offer free detection tools that allow users to assess whether content was AI-generated.
• Disclosure to downstream users — Deployers (anyone using AI tools to create content for distribution) must disclose AI involvement.

SB 942 explicitly names C2PA as the kind of “widely adopted industry standard” it envisions. While IPTC Digital Source Type is also a standard, the legislation's emphasis on provenance data that supports detection suggests that unsigned metadata alone may not satisfy the spirit of the law. For more detail, see our full SB 942 compliance guide.

Action Plan: Preparing for August 2026

You have approximately five months. Here is what to do with them.

1. Audit Your Published Midjourney Content

Start by understanding your exposure. Identify all Midjourney-generated content that is currently published, shared with clients, or distributed to European audiences. For each piece, check:

• Does the published version retain the IPTC Digital Source Type tag? (Most social media platforms will have stripped it.)
• Is there a human-readable AI disclosure label alongside the content?
• Can you trace the image back to its original Midjourney generation? (Do you have the Job ID, the original file, or generation records?)

2. Implement Disclosure Workflows

Do not wait for Midjourney to add C2PA. Build the workflow yourself:

• Preserve the IPTC tag — It is already there. Ensure your editing and export workflows do not strip it. Test your Photoshop, Figma, and CMS export settings.
• Add human-readable labels — Include “Generated with AI” or equivalent disclosures alongside AI-generated content in client deliverables, website publications, and marketing materials.
• Maintain an audit trail — Keep records linking published content to its AI origin: Job IDs, generation dates, original prompts, and the disclosure measures applied. If a regulator asks “was this AI-generated and was it disclosed?” you need to answer both questions.

3. Understand Jurisdiction-Specific Requirements

Article 50 is the headline, but implementation details will vary. Key questions to monitor:

• Will the European Commission's implementing acts specify particular technical standards (C2PA, watermarking, or both)?
• Do your clients operate in California (SB 942), the EU (AI Act), or both? Requirements may differ.
• Does your industry have sector-specific guidance? Advertising, news media, and entertainment may have additional disclosure norms.

For a comprehensive jurisdiction-by-jurisdiction view, see our global AI content disclosure laws overview and the AI art compliance checklist.

4. Prepare Client Communication

If you produce AI-generated content for clients, you need to communicate proactively:

• Which deliverables used AI generation (and which did not)
• What disclosure measures are in place (IPTC tags, human-readable labels, audit trails)
• What additional steps may be needed when implementing acts provide further technical guidance
• How your workflow handles the gap between IPTC (present) and C2PA (not yet available from Midjourney)

Transparency with clients now is far better than a surprise compliance conversation in August. For contract-level guidance, see our article on AI clauses in agency contracts.

The Disclosure Gap Is a Workflow Problem

Midjourney has made a genuine effort with IPTC Digital Source Type. It is a real standard, it is embedded consistently, and it gives compliance teams something to work with. But unsigned metadata in a world of metadata-stripping platforms is a disclosure that does not survive contact with the real internet.

The gap between “present at creation” and “detectable at consumption” is not a Midjourney problem alone—it is a workflow problem. Until Midjourney adds C2PA (or until the EU clarifies its technical standards), the responsibility falls on deployers: the agencies, freelancers, and brands who use AI-generated content and publish it for European audiences.

August 2026 is five months away. The regulation is published. The deadline is fixed. The metadata is partially there. The question is whether your workflow closes the rest of the gap.