Both questions hit on the key risk of commercial AI image use:
1. IP warranty: You SHOULD NOT warrant that AI-generated images are "original" or "non-infringing" without qualification. You can't fully verify this. Instead, offer a modified warranty: "To the best of Agency's knowledge, deliverables do not knowingly infringe third-party intellectual property." This is honest โ you're not guaranteeing what the AI model may have been trained on.
2. Infringement liability: If a DALL-E output substantially resembles a copyrighted work, the liability chain is unclear. OpenAI's terms disclaim most liability. The client could be liable for using it. And your agency could be liable for delivering it. In practice, the risk is low for most commercial use (generic marketing images), but increases for specific styles, characters, or recognizable compositions.
Risk mitigation: (1) run reverse image searches on AI outputs before delivery, (2) disclose to clients that AI tools were used (some clients require this), (3) maintain E&O insurance that covers AI-generated content, (4) use AI outputs as starting points and add substantial human modifications.