1.1 AI Services Covered

This addendum applies to the following artificial intelligence services:

1.2 Key Terminology

"AI Model"

A machine learning algorithm trained to make predictions, classifications, or decisions based on data.

"Training Data"

Data sets used to train AI models, including both Client-provided and third-party data sources.

"AI Output"

Results, predictions, recommendations, or content generated by AI systems.

"Bias"

Systematic errors or unfair discrimination in AI outputs that may affect certain groups or individuals.

2.1 Accuracy and Reliability

• AI models provide predictions and recommendations, not guarantees
• Accuracy may vary based on data quality, model complexity, and use case
• Performance may degrade over time as data patterns change
• Results may not be suitable for all business decisions or contexts
• Human oversight and validation remain essential for critical decisions

2.2 Bias and Fairness

AI systems may exhibit bias due to:

• Historical bias present in training data
• Sampling bias or incomplete data representation
• Algorithmic bias in model design or implementation
• Evolving social and cultural standards

Anthromorphe implements bias mitigation strategies but cannot guarantee completely bias-free results.

2.3 Data Dependencies

AI performance depends on:

• Quality, completeness, and accuracy of training data
• Relevance of historical data to current conditions
• Ongoing data quality and consistency
• Sufficient data volume for statistical validity

2.4 Explainability Limitations

Some AI models, particularly deep learning systems, may operate as "black boxes" where the decision-making process is not fully explainable or interpretable by humans.

3.1 Data Responsibilities

Client is responsible for:

• Ensuring data accuracy, completeness, and legal compliance
• Obtaining necessary permissions for data use and processing
• Maintaining data security and access controls
• Providing representative and unbiased training datasets
• Regular data quality monitoring and maintenance

3.2 Human Oversight Requirements

3.3 Appropriate Use Guidelines

Client agrees to use AI systems:

• In accordance with applicable laws and regulations
• With appropriate disclaimers about AI-generated content
• With consideration for fairness and non-discrimination
• With regular monitoring for performance degradation
• With respect for privacy and data protection rights

3.4 Testing and Validation

Client must thoroughly test AI systems before deployment and implement ongoing monitoring to ensure continued performance within acceptable parameters.

4.1 Training Data Usage

Client-provided data will be used solely for developing and training AI models for Client's specific project unless otherwise agreed in writing.

4.2 Data Retention and Disposal

Training and processing data will be:

• Retained only as long as necessary for model development and support
• Securely deleted upon project completion or Client request
• Stored with appropriate security measures during engagement
• Not used for other projects without explicit Client consent

4.3 Third-Party Data Sources

When third-party data is incorporated, Client acknowledges:

• Compliance with third-party data licensing terms is required
• Data quality and bias may be outside our direct control
• Ongoing licensing fees may apply
• Data availability and access may change over time

4.4 Synthetic and Generated Data

AI-generated synthetic data created during development becomes Client property upon full payment, subject to applicable intellectual property laws.

5.1 Regulatory Landscape

AI regulations are rapidly evolving. Client is responsible for ensuring compliance with applicable laws including:

• EU AI Act and related European regulations
• US federal and state AI governance frameworks
• Industry-specific regulations (healthcare, finance, etc.)
• Data protection and privacy laws (GDPR, CCPA, etc.)
• Anti-discrimination and civil rights laws

5.2 Ethical AI Principles

Anthromorphe follows these ethical AI principles:

• Transparency: Clear documentation of AI capabilities and limitations
• Fairness: Efforts to minimize bias and promote equitable outcomes
• Accountability: Clear responsibility for AI system behavior
• Privacy: Respect for individual privacy and data rights
• Human-Centric: AI should augment, not replace, human judgment

5.3 Risk Assessment

High-risk AI applications require additional considerations including impact assessments, enhanced testing, and specialized documentation.

5.4 Compliance Monitoring

Client is responsible for ongoing monitoring of AI systems to ensure continued compliance with evolving regulations and internal policies.

6.1 Performance Metrics

AI system performance will be evaluated using:

• Accuracy, precision, recall, and F1 scores where applicable
• Business-specific KPIs defined in project requirements
• Model drift detection and performance monitoring
• User satisfaction and adoption metrics

6.2 Model Maintenance

Ongoing model maintenance may include:

• Regular retraining with updated data
• Performance monitoring and alerting
• Bug fixes and security updates
• Feature updates and enhancements

6.3 Support Limitations

Support does not include:

• Guarantee of specific performance levels
• Customization for regulatory changes
• Data quality issues originating from Client systems
• Integration issues with unsupported third-party systems

6.4 End-of-Life Planning

Plans for model retirement, data migration, and system replacement should be considered as part of the initial project planning.

7.1 No Guarantee of Outcomes

Anthromorphe does not warrant that AI systems will:

• Achieve specific accuracy or performance levels
• Produce consistently reliable results
• Comply with future regulatory requirements
• Eliminate all bias or unfair discrimination
• Replace the need for human judgment and oversight

7.2 Third-Party AI Services

When integrating third-party AI services (OpenAI, Google AI, etc.), Anthromorphe disclaims responsibility for:

• Service availability and performance
• Changes to APIs, pricing, or terms of service
• Data handling practices of third-party providers
• Content generated by third-party AI models

7.3 Enhanced Liability Limitations

In addition to general liability limitations, Anthromorphe is not liable for damages arising from AI-specific risks including algorithmic bias, model drift, or regulatory non-compliance.