CHILE Law and Practice Contributed by: Carolina Cabrera, LawTech
access, location, subcontracting, transfers, deletion, anonymisation and security. Customers also request transparency, documentation and auditability, including capabilities and limitations, model versioning, testing, data sources at a high level, accuracy, bias controls, human oversight, logs and explainability. For higher-risk uses, impact assess - ments and risk management may be required. IP, security and output risk are heavily negotiated. Customers seek clarity on prompts, outputs, fine- tuned models, embeddings, datasets and bespoke developments; indemnities for third-party IP claims; safeguards against hallucinations, bias, harmful con - tent and prompt injection; and future-proofing clauses for regulatory change. 3.2.2 AI-Related Liabilities in Contracts AI-related liability in Chile is currently addressed main - ly by contract because there is not yet a fully enacted general AI liability regime. Parties allocate risk through warranties, disclaimers, indemnities, liability caps, carve-outs, compliance duties, human review, data protection, cybersecurity, IP and incident response clauses. A common approach is to distinguish liability for the AI tool from liability for the customer’s use of it. Sup - pliers state that outputs may be probabilistic or inac - curate and should be reviewed. Customers seek com - mitments that the system is tested, documented and configured for the agreed purpose. Generative AI contracts increasingly address halluci - nations, inaccurate outputs and customer decisions based on outputs. Customers try to narrow disclaim - ers where the supplier made specific performance claims, provided a bespoke or fine-tuned model, con - trolled the workflow or marketed the tool for a regu - lated or high-impact use. Data and IP liability are major negotiation points. Contracts allocate responsibility for lawful input data, prompts, training data and fine-tuning datasets. Cus - tomers seek protection against unauthorised training, disclosure, unlawful transfers, security incidents and third-party IP claims.
Liability caps are becoming more differentiated. Sup - pliers seek aggregate caps and exclusions, while customers request carve-outs or higher caps for con - fidentiality, data protection, IP, cybersecurity, fraud, wilful misconduct, gross negligence, regulatory fines caused by the supplier and failures affecting critical services. 3.3 Key Concerns of Providers AI providers in Chile are mainly concerned with avoid - ing an excessive transfer of risk, especially while the AI regime and market standards are still evolving. Their concerns usually relate to outputs, customer misuse, data rights, IP, confidentiality, cybersecurity, auditability and future regulatory change. A key concern is liability for outputs. Providers resist full responsibility for hallucinations, inaccurate or biased results, or decisions made by the customer where the customer controls the use case, inputs, prompts, configuration, workflow or final decision. They usually require human review and acceptable- use restrictions. Providers are also concerned about misuse, including unlawful, discriminatory, infringing, abusive or high- risk uses outside scope, scraping, reverse engineer - ing, model extraction, prompt injection, bypassing safety controls, or uses affecting third parties or other customers. Data and IP are heavily negotiated. Providers want customers to be responsible for the lawfulness and quality of inputs, prompts, training data and fine-tun - ing datasets. They also seek to preserve ownership of models, algorithms, pre-existing technology, docu - mentation, know-how, embeddings, generic improve - ments and platform developments. Transparency, audit and cybersecurity requests are another concern. Providers must balance customer and regulator demands against confidentiality, trade secrets, security and obligations to other customers. They also seek suspension, investigation, coopera - tion and change-control rights where customer use creates platform or compliance risk.
15 CHAMBERS.COM
Powered by FlippingBook