AI LIABILITY ADDENDUM

PREAMBLE

This AI Liability Addendum to Main Service Agreement ("Addendum") for [Service Name] dated [Date] between [Provider] and [Customer] establishes liability allocation for AI system errors/failures. Legal Framework: EU AI Act 2023/1230 Art. 28-31 (Liability), US UCC §2-102 (Goods Warranty), UCC §2-313 (Express Warranty), German BGB §280 (Liability).

1.1 AI System Limitations (EXPLICIT DISCLAIMER): Provider makes NO guarantee regarding AI system per UCC §2-316 (Disclaimer):
☑ Accuracy: AI outputs may contain errors (accuracy NOT guaranteed)
☑ Completeness: AI may miss information OR provide irrelevant data
☑ Reliability: AI may fail/malfunction without notice
☑ Fitness: AI suitable only for informational purposes (NOT decision-making)
CRITICAL LANGUAGE: "AI OUTPUTS ARE PROVIDED 'AS-IS' WITHOUT WARRANTY OF ANY KIND."

1.2 AI Cannot Replace Human Judgment: Provider explicitly disclaims per AI Act Art. 28(1) (Responsibility):
• AI is NOT substitute for professional advice (legal, medical, financial)
• Human must verify ALL AI outputs before use/reliance
• AI errors may have serious consequences (financial loss, harm)
1.3 No Liability for AI Errors: Provider NOT liable per UCC §2-316(2) for:
✗ Inaccurate AI responses
✗ Hallucinations (AI-invented information)
✗ Bias/discriminatory outputs
✗ Outdated information (training data limitations)
✗ Harmful advice based on AI (user responsibility to validate)

2.1 Customer Responsibility (BINDING): Customer acknowledges per German BGB §311 (Duty of Care):
☑ AI is supplementary tool (not final authority)
☑ Customer responsible for verifying AI outputs (accuracy check required)
☑ Customer must not rely solely on AI for critical decisions
☑ Customer must implement human review process (per AI Act Art. 6(1) requirements)
☑ Customer accepts risk of AI errors (assumption of risk)
2.2 Validation Obligations: Before using AI output, Customer must per AI Act Art. 28(1):
• Review accuracy (especially for financial/legal/medical decisions)
• Cross-check against multiple sources (if material consequence)
• Verify reasoning (ensure AI logic sound)
• Test edge cases (does AI handle unusual scenarios?)
2.3 Prohibited Reliance: Customer shall NOT use AI for (without explicit Provider approval):
✗ Life-critical decisions (medical diagnosis, emergency response)
✗ Legal judgment (contract interpretation, compliance advice)
✗ Financial advice (investment recommendations, loan decisions)
✗ Criminal justice applications (risk assessment, sentencing)
• Using AI in prohibited areas = Customer assumes ALL liability (per AI Act Art. 28(2))

3.1 Provider Liability (LIMITED): Provider liable ONLY for per UCC §2-719 (Limited Remedy):
☑ Direct damages (actual out-of-pocket cost)
☑ NOT for lost profits, lost opportunity, business interruption
☑ Liability cap: Lesser of (a) fees paid in [12 months] OR (b) direct damages
☑ Exception: Liability NOT limited for (a) fraud, (b) gross negligence, (c) willful misconduct
3.2 AI Error Liability (SPECIFIC): If AI error occurs per German BGB §280(1):
• Provider NOT liable if Customer failed due diligence (verification responsibility)
• Provider liable ONLY if error caused by Provider negligence (system design flaw)
• Standard: Provider liable if error foreseeable + preventable (duty of care test)
3.3 Customer Indemnification (BINDING): Customer indemnifies Provider per UCC §2-102 from:
• Third-party claims arising from Customer's use of AI outputs
• Liability for damages from Customer's reliance on AI (without validation)
• Regulatory fines (if Customer used AI in violation of law)

4.1 AI Bias (ACKNOWLEDGED LIMITATION): Provider acknowledges per AI Act Art. 6(1) (High-Risk) that AI may exhibit bias:
☑ Bias sources: Training data, model design, parameter tuning
☑ Acknowledged biases: [if known, list: e.g., "system may underestimate risk for certain demographics"]
☑ Mitigation: Provider implements fairness testing (per [X frequency])
☑ NO liability: Provider NOT liable for bias-related errors (bias inherent to AI per AI Act Art. 26)
4.2 Customer Mitigation Obligation: Customer responsible for per AI Act Art. 28(1):
• Monitoring AI for bias (track outcomes across demographics)
• Human review of high-stakes decisions (hiring, credit, justice)
• Documenting bias concerns (complaint log required)
4.3 Discriminatory Output Liability: If AI outputs violate anti-discrimination law per EU Gender Equality Directive:
• Customer liable (not Provider) if Customer uses without human review
• Provider liable ONLY if bias was known + undisclosed (fraud)
• Both parties may face regulatory action (joint liability possible)

5.1 AI-Generated IP Ownership: Per 17 USC §201 (Copyright Ownership):
☐ Provider owns: AI-generated content (if solely AI-created)
☐ Customer owns: AI-generated content produced for Customer (work-for-hire)
☐ Specify: [who owns AI-generated outputs]
5.2 Training Data Concerns: If AI trained on copyrighted data per 17 USC §107 (Fair Use):
• Fair use defense: AI training may qualify as fair use (courts determining)
• Copyright infringement risk: AI outputs may infringe on training data sources
• Provider liability: LIMITED (fair use doctrine protects Provider in most cases)
5.3 Customer Responsibility (IP Risk): Customer responsible for:
• Verifying AI outputs don't infringe third-party IP
• Obtaining licenses if AI output uses copyrighted material
• Indemnifying Provider if Customer publishes AI outputs that infringe

6.1 AI Act Compliance (JOINT RESPONSIBILITY): Per AI Act Art. 28:
• Provider: Implements technical measures per Art. 28(2)
• Customer: Responsible for use case compliance (not using AI for high-risk without safeguards per Art. 6)
• Risk: Regulatory fines up to EUR 30M / 6% global revenue (Art. 71)
6.2 Regulatory Fines (ALLOCATION): If regulatory action taken per AI Act Art. 71-75:
• Customer-caused violation: Customer liable (used AI outside approved scope)
• Provider-caused violation: Provider liable (system design flaw)
• Shared responsibility: Both parties proportional to fault
6.3 Certification & Compliance Documentation:
• Provider maintains: Compliance records (AI system design, fairness testing)
• Customer maintains: Usage records (decisions made, human review applied)
• Both parties cooperate: With regulators/auditors on request

7.1 Dispute Resolution: AI liability disputes resolved per AI Act Art. 85 (Dispute Settlement):
• Mediation: [30 days] to resolve in good faith
• If unresolved: Binding arbitration (per main agreement)
• Burden of proof: Provider must show Customer failed due diligence (to avoid liability)
7.2 Waiver of Implied Warranties: Customer waives per UCC §2-316:
☑ Implied warranty of merchantability (AI not warranted for any purpose)
☑ Implied warranty of fitness (AI not fit for particular purpose)
☑ Any other implied warranties
ACKNOWLEDGMENT (SIGNATURE REQUIRED):
"Customer acknowledges that AI outputs may be inaccurate/harmful and accepts the risk of using AI without human validation."
Customer Signature: _____________ Date: _____________

Law: ☐ German (BGB + AI Act) ☐ [US (UCC + State Law)] | Enforcement: AI Act violations subject to regulatory fines (EUR 30M / 6% revenue) + private claims for damages