AI Training & Model Use

01 The short version, restated

The risk we're addressing here: when you paste text or upload an image into an AI tool, some vendors may use that input to train their models. Once your material is in a training set, it's effectively unrecoverable. We don't put your work into tools that do that.

02 What we ask of every AI vendor we use

• Their terms must let us turn off training on our inputs (or training must not be the default for business accounts).
• They must give us a written data-handling commitment we can rely on.
• They must not retain inputs longer than needed for the immediate task.

03 Vendor-by-vendor position

OpenAI (ChatGPT, GPT API)

• API inputs are not used to train models by default.
• ChatGPT business and enterprise tiers do not train on input.
• We use business-tier accounts for any client-related work.

Anthropic (Claude)

• Claude does not train on inputs from API or business-tier users by default.
• We use business-tier accounts.

Adobe (Firefly, Generative Fill)

• Firefly is trained on Adobe Stock and licensed content. Adobe's commercial terms cover output usage.
• Generative Fill in Photoshop does not, per current terms, train on user inputs.

Canva (Magic Studio)

• Canva's terms commit to not using customer designs to train AI without permission. We rely on that commitment.

Lovable

• Lovable is the platform powering cmykink.io. Per Lovable's published terms, customer project content is not used to train AI models. Note for legal review: this requires a fresh check against Lovable's current ToS at the time of publication.

Otter.ai

• Otter.ai uses transcripts to improve transcription accuracy. We use the business tier; we delete sensitive recordings promptly; and we don't use Otter for transcribing material covered by NDA without participant agreement.

04 What this means in practice

• Your project files, briefs, and creative direction are not used to train any AI model we know of.
• Where a tool's training position is ambiguous or worse, we don't put your work into it.
• When we use AI in your project, prompts and outputs live on the vendor's servers under their data-handling commitments, not in a training pipeline.

05 Your opt-out

If you don't want any AI tool involved in your project — even tools that don't train on input — tell us. We honor it. See Policy 06 for the broader AI opt-out.

06 Enterprise clients

If your organization has stricter requirements — explicit no-AI workflows, specific vendor restrictions, security questionnaires — we can usually accommodate them. Bring it up before the project starts so we can scope correctly.

07 Changes and review

Vendor terms change. This policy is reviewed at least quarterly and whenever a vendor announces a material change to its training practices.