Source: JD Supra

AI-generated outputs raise unresolved copyright and ownership questions where human authorship is limited, with guidance from the U.S. Copyright Office and recent cases indicating that purely AI-created works lacking meaningful human input may not be eligible for registration and that prompts alone are generally insufficient to confer authorship. Contracts should therefore contain explicit ownership clauses allocating rights in outputs and derivatives, and parties must review upstream model-provider terms because those terms constrain the rights a technology company can grant to its downstream customers. Some model vendors expressly assign or condition ownership to customers, while others retain rights; training and data-use provisions—such as vendor rights to use customer inputs for system improvement or to retain aggregated usage data—can create confidentiality and trade-secret risks if not tightly circumscribed.

Indemnification practices and insurance coverage for output-related IP risk vary widely: some providers disclaim indemnities for customer use, others offer limited or carved-out indemnities, and many vendors resist broad representations about training-practice compliance, often qualifying them by knowledge or materiality. Parties should align upstream indemnity rights with downstream commitments to avoid gaps, implement practical safeguards (human review, content filters, similarity detection, contemporaneous records of human involvement), and include contractual data-use, audit and transparency rights to address potential processing of personal data and related controller/processor issues. Contracts should also require disclaimers and human oversight to mitigate risks from biased or incorrect outputs, and parties must monitor evolving U.S. state-level AI laws that affect disclosure and provenance obligations while balancing those obligations against protection of proprietary information.