DALL·E 3
OpenAI's text-to-image model integrated into ChatGPT and available via API. Generates high-fidelity images from natural-language prompts with strong text rendering and prompt adherence.
Independent assessment across data handling, compliance, security and transparency.
Overview
DALL-E 3 is OpenAI's third-generation text-to-image model, available to consumers inside ChatGPT Free, Plus, and Pro, and to developers through the OpenAI Images API. It excels at rendering legible text within images, following long compositional prompts, and producing on-brand marketing and illustration assets at 1024x1024 up through 1792x1024 resolutions. For enterprise risk, the default consumer tier (ChatGPT Free/Plus) has model training enabled by default and must be toggled off under Data Controls. API and ChatGPT Enterprise/Team usage is excluded from training by contract. Generated images carry C2PA provenance metadata. Copyright indemnity is offered only on enterprise tiers, so shadow-AI use on personal accounts exposes employees to IP and confidentiality risks when they paste proprietary references.
Risk factors
3- Integrated into ChatGPT, potentially sharing data.
- Generates images based on user prompts.
- No clear opt-out for training on user data.
Recommendations
8- Standardize on ChatGPT Enterprise or Team (training off by default, DPA, SOC 2)
- Disable 'Improve the model for everyone' toggle under Data Controls on all consumer accounts
- Block uploads of logos, unreleased designs, customer photos and PHI via DLP
- Require C2PA metadata to remain attached on any externally published AI image
- Route production workflows to the OpenAI API with Zero Data Retention where available
- Publish an AI-generated content usage policy covering IP, attribution, and brand-safe prompts
- Train marketing and design teams on copyright risk of style mimicry prompts
- Log DALL-E usage via egress and browser telemetry to detect shadow-AI patterns
Data handling
- Storage
- US-based OpenAI infrastructure on Azure; 30-day abuse-monitoring retention even when training is disabled.
- Retention
- Conversations retained until user deletion; deleted chats purged within 30 days; abuse logs kept 30 days minimum.
- Training on inputs
- Consumer ChatGPT Free/Plus: prompts and images used to train models by default; user must opt out. API, Team, Enterprise: excluded from training by default.
More Image Gen tools
See all →Adobe Firefly
Adobe Firefly is a commercial-safe generative AI suite for images, video, and vector assets, trained on licensed Adobe Stock and public-domain content with IP indemnification for enterprise customers.
Adobe Stock AI
AI-generated images within Adobe Stock marketplace.
Appen AI
Data annotation and AI training data platform.
Artbreeder
Collaborative AI art tool using genetic image blending.
Artsmart AI
AI image generator for stock photos and marketing visuals.
Astria AI
AI platform for fine-tuning and generating custom product images.