This is AI21 Labs' legal agreement that governs how you can use their AI services, including models like Jamba and their Studio API platform. The most important thing to know is that AI21 Labs may use the text you input into their services to improve their AI models unless you have a separate enterprise agreement that says otherwise. If you are a business or developer using the API, you should review whether you have signed a data processing agreement and whether your enterprise contract explicitly restricts AI21 from training on your data.
This document governs the use of AI21 Labs' AI-powered products and services, including its APIs, developer tools, and consumer-facing applications such as Jamba models, and is legally structured as a binding contract formed upon user acceptance. Key obligations include users' responsibility to comply with acceptable use policies, prohibition on using outputs to train competing AI models, and AI21's right to modify, suspend, or terminate services at its sole discretion. Notable deviations from industry standard include a broad intellectual property assignment clause covering user inputs and outputs in certain contexts, AI21's explicit reservation of the right to use input data for model improvement absent a contrary enterprise agreement, and a limitation of liability cap set at amounts paid in the prior 12 months or $100 (whichever is greater), which is exceptionally low for AI infrastructure providers. The document engages GDPR (for EU/EEA users), CCPA (for California residents), and the emerging EU AI Act framework, and raises material compliance considerations around data processing agreements, cross-border data transfers, and the lawfulness of using customer data to train AI models without granular consent.
🔒 Institutional analysis locked
Regulatory exposure by statute, material risk assessment, vendor due diligence action items, and enforcement precedent. Available on Professional.
Upgrade to Professional — $149/mo1 change analyzed since monitoring began.
Severity increased from high to high, but the provision text remains identical; the actual change is the severity rating elevation from medium to high.
9 provisions unchanged.
View full change record →Cross-platform context
See how other platforms handle Acceptable Use Policy and Content Restrictions and similar clauses.
Compare across platforms →