Prohibition on AI/ML Training Use of Service Content

What it is

You are not allowed to use any content from the Fireworks AI platform, including metadata and outputs, to train or develop your own AI or machine learning models.

Why it matters (compliance & governance perspective)

For developers and companies building AI-enabled products on top of Fireworks AI, this restriction may limit how outputs can be used in downstream model training pipelines, which is a core use case for many API customers.

Consumer impact (what this means for users)

Developers and organizations that use Fireworks AI outputs to train or fine-tune their own AI models may be in violation of this clause, even if those downstream uses occur outside the Fireworks platform itself.

Institutional analysis (Compliance & governance intelligence)

(1) REGULATORY LANDSCAPE: This restriction is a contractual use limitation rather than a regulatory requirement, but it interacts with emerging AI governance frameworks including the EU AI Act, which imposes transparency and data governance obligations on AI system developers. The clause also engages general IP law principles around the permissibility of using outputs from a third-party service for derivative development. The FTC has begun examining AI training data practices under its consumer protection and competition authority. (2) GOVERNANCE EXPOSURE: Medium to High for AI-focused enterprise customers. The restriction explicitly prohibits use of Service content, including metadata and outputs, for ML/AI training and for identification of natural persons. In the AI industry context where inference outputs are frequently used for model distillation, fine-tuning, or synthetic data generation, this clause directly restricts a commercially significant downstream use case. (3) JURISDICTION FLAGS: EU/EEA users may find this clause interacts with the EU AI Act's requirements around training data documentation and prohibited AI practices, particularly the restriction on technologies for identification of natural persons. The enforceability of output-use restrictions under copyright and contract law varies by jurisdiction; some courts may view this as an unenforceable restraint on the use of non-copyrightable outputs. (4) CONTRACT AND VENDOR IMPLICATIONS: AI product teams should carefully assess whether their intended use of Fireworks AI outputs for fine-tuning, reinforcement learning from human feedback, or distillation workflows is prohibited by this clause. Vendor assessments should flag this restriction for legal review before building production pipelines that incorporate Fireworks outputs in training datasets. (5) COMPLIANCE CONSIDERATIONS: Organizations should audit their AI development pipelines to determine whether Fireworks AI outputs are being used in training contexts. If so, legal teams should assess whether a contractual amendment or separate licensing arrangement with Fireworks is required. The restriction on technologies for identification of natural persons may also implicate biometric data laws in Illinois (BIPA) and other jurisdictions if facial recognition or similar applications are contemplated.

Applicable agencies

Provision details

Other risks in this policy

• Mandatory Individual Arbitration high
• Class Action and Jury Trial Waiver high
• Organizational Account Liability high
• Fireworks IP License Over User Content medium
• Unilateral Termination Without Cause medium
• Minor Access Restrictions medium

Related Analysis

• Robinhood's Agreement Authorizes Position Liquidation Without Prior Notice

  Buried in Robinhood's customer agreement is broad authority to close your positions, suspend your account, and force arbitration. Here is what it actually says.

• Stripe's Reserve and Hold Authority: What the Terms Authorize

  Stripe's terms authorize fund reserves, payout withholding, and account termination. Here is what the agreement states and what business owners should review.