共有:
AI Copyright × Licensing Deals

OpenAI-Getty Deal Reduces AI Copyright Risk

AI Navigate Editorial / 2026-06-24 / 6 min read

The legitimacy of training data for AI image generation has remained unresolved since the Getty lawsuit in 2023.

2023 Getty Lawsuit High Risk 2024 Partial Settlements Mid Risk 2026 Getty Formal Deal Risk Reduced Ongoing NYT & others Unresolved Hi Mid Lo

01 — The Origin: 2023 Getty vs. Stability AI

In January 2023, Getty Images filed suit against Stability AI in both UK and US courts, alleging that over 12 million images had been used without authorization as training data. The complaint included screenshots of AI-generated images still bearing Getty watermarks — sending shockwaves across the entire AI image generation industry.

The lawsuit went beyond a dispute between two companies. It put the question of training data legality for generative AI squarely before the industry. OpenAI, Midjourney, Adobe Firefly, and every other major service was reported to face similar risks, prompting both investors and enterprise users to grow increasingly wary of legal uncertainty.

"The generation of watermarked images is irrefutable evidence of the origin of the training data."

Summarized from Getty Images complaint (2023)
FiledJanuary 17, 2023
DefendantStability AI Ltd.
ClaimsCopyright infringement & trademark dilution
Images at issueOver 12 million

02 — What the Formal License Agreement Covers

OpenAI has entered into a formal licensing agreement with Getty Images, progressively eliminating copyright risks around training data and image generation for GPT Image 2.0 and Sora 2.

1
Training Data License

Getty-owned photos and video content are now authorized for model training in GPT Image 2.0 and Sora 2. The arrangement includes a revenue-sharing scheme for creators whose work is used.

2
Clean Title for Generated Outputs

Images and video generated from licensed data carry significantly reduced copyright claim risk for commercial use. Both parties have mutually acknowledged indemnity provisions.

3
Creator Royalty Model

A royalty mechanism activates each time a Getty-registered creator's work is used in training. A portion of AI revenue is returned to content contributors.

4
Audit Rights and Transparency Reports

Annual audits of training data usage are permitted, with Getty authorized to appoint an independent auditor. OpenAI is obligated to publish utilization reports every six months.

Models Covered

GPT Image 2.0, Sora 2 (as of June 2026)

Scope

Includes commercial use via API and product integrations

Exclusions

Some features in ChatGPT free tier; research and educational use governed by separate terms

03 — Risks That Remain: NYT Lawsuit and Ongoing Cases

The Getty agreement is an important step, but it does not resolve the industry's copyright issues as a whole.

Risks Eliminated

  • Use of Getty-owned images as training data
  • Trademark dilution claims on GPT Image 2.0 / Sora 2 outputs
  • Watermarked-output problem (Getty content)
  • Criticism over training data opacity (Getty content)

Still Unresolved

  • NYT vs. OpenAI — text copyright, litigation ongoing
  • Authors Guild and other book-copyright suits
  • Separate copyright suits against Midjourney and Stability AI
  • Transparency obligations under the EU AI Act (effective 2026)
  • Unlicensed non-Getty media companies
The NYT lawsuit (filed December 2023) concerns text copyright in a separate legal arena from image generation. If any court awards significant damages, the impact on the broader industry could be unavoidable — a risk that warrants close monitoring.

04 — Practical Guidance for Teams Using Image Generation APIs

Legal risk for teams using image generation APIs has dropped somewhat. With suits like the NYT case still active, this is not a complete resolution — but here is what you can do right now.

A
Verify the License Status of Your API

If you are using OpenAI's GPT Image 2.0 or Sora 2, you benefit from the Getty agreement. Midjourney and Stability AI-based services require separate verification. Always review the scope of the "indemnity clause" in your provider's terms of service.

B
Document Your Generated Outputs

Log which model and version produced each output, and when. This record becomes critical evidence in any future litigation or compliance audit.

C
Manage Text Generation Risk Separately

As the NYT lawsuit illustrates, text copyright risk remains unresolved. Use cases that prompt the model to reproduce news articles or book passages verbatim carry elevated exposure. Consider managing sources explicitly through a RAG architecture.

D
Schedule Regular Reviews with Legal

Copyright verdicts and settlements will keep coming. Establish a cycle — at minimum every six months — to review the licensing status of your generative AI services and track relevant case law with your legal team.


Summary: The formal licensing agreement with Getty Images is the first significant settlement in the generative AI copyright space. However, with the NYT lawsuit, book-copyright cases, and the EU AI Act still in play across multiple fronts, this is best understood as a "gradual reduction of risk" rather than a "complete resolution." Other AI providers are expected to accelerate their own licensing negotiations in OpenAI's wake.