Generic SaaS terms don't address output ownership, training data rights, or hallucination disclaimers. I check for the provisions that matter when you're shipping AI.
Beyond standard SaaS - these are what make AI terms bulletproof
Who owns AI-generated content? What license do users get for outputs? Can you use outputs to improve your model? Customer indemnification for how they use outputs?
Do you train on customer inputs? Is there an opt-out mechanism? How is data anonymized? Can competitors' data train your model? This is make-or-break for enterprise deals.
No guarantee of accuracy. Human review requirements. Prohibited use cases (medical, legal, financial advice). Liability cap for reliance on outputs.
How long are prompts/inputs retained? Are inputs logged, and where? Confidentiality of customer prompts? Input data deletion rights?
Notice before model updates? Ability to pin to specific model version? Deprecation timeline for older models? Performance consistency guarantees?
Token/request limits clearly defined? Overage pricing transparent? Throttling policies? Fair use restrictions that could catch enterprise users off guard?
AI-specific provisions layer on top of these SaaS essentials
Service availability commitments
Customer data belongs to customer
Limitation of damages
IP and third-party claims
Choice of law & venue
Exit rights & data return
SOC 2, encryption, breach notice
CA compliance disclosure
Paste your Terms of Service below. I check AI-specific provisions first, then standard SaaS clauses.
Want the complete picture? My full attorney review catches nuanced issues this scanner can't detect - IP ownership edge cases, liability carveouts, and enforceability risks specific to AI products.
Get Full Attorney Review — $349