Silvia AI and Algorithmic Transparency Statement

Last Updated: September 2026

1. Our Commitment to Transparency

Silvia, Inc. (“Silvia,” “we,” “us,” or “our”) builds an AI-powered personal finance platform that helps people understand and manage their financial lives. Artificial intelligence is central to the Services we provide. We believe that people who entrust us with their financial data deserve a clear, honest explanation of how our AI works, what it can and cannot do, and how we govern it.

This AI and Algorithmic Transparency Statement (“Statement”) describes the AI technologies used in Silvia, the data that trains and powers them, the safeguards in place, the limitations you should understand, and the choices available to you. This Statement supplements our Privacy Policy, the Terms of Service, and the Acceptable Use Policy.

2. AI Disclosure

When you use Silvia, you are interacting with an AI system. The financial analyses, projections, recommendations, categorizations, summaries, alerts, and conversational responses delivered through the Services are generated, in whole or in part, by artificial intelligence and machine learning models. Silvia is not a human financial advisor, and no human reviews every individual Output before it is delivered to you.

This disclosure is provided in accordance with emerging transparency frameworks, including Article 50 of the EU AI Act, which requires that individuals be informed when they are interacting with an AI system.

3. How Our AI Works

3.1 AI Capabilities

3.2 Types of AI Models

• Large Language Models (LLMs): Pre-trained foundation models that generate natural language responses, summaries, and explanations. Developed by third-party providers, including open source models that may be fine-tuned by Silvia.
• Classification Models: Trained to categorize transactions, detect anomalies, and assign labels. Trained on aggregated, anonymized data.
• Statistical and Forecasting Models: Time-series and regression models for projections, trend analyses, and scenario modeling.
• Rule-Based Systems: Deterministic algorithms applying defined formulas and thresholds. Not AI in the strict sense but part of the algorithmic infrastructure.

3.3 How Decisions Are Made

Most AI-generated Outputs are informational, not decisional. Silvia does not make financial decisions on your behalf. No AI Output is used for automated credit scoring, approval or denial of financial products, automated trading without your explicit instruction, or profiling that produces legal or similarly significant effects.

If Silvia introduces any feature involving automated decision-making with legal or significant effects, we will update this Statement, provide specific disclosure, and implement safeguards including the right to human review.

4. Data Used to Train and Power AI

4.1 Training Data

• Anonymized User Data: De-identified data derived from user Inputs, Outputs, and transaction records. All direct identifiers removed and quasi-identifiers suppressed before use.
• Publicly Available Data: Financial knowledge and general information incorporated into foundation models during pre-training by third-party providers.
• Licensed Data Sets: Commercial data licensed for specific improvements such as transaction categorization accuracy.
• Feedback Signals: Aggregated, de-identified signals from thumbs-up/down, categorization corrections, and explicit feedback.
• Synthetic Data: Artificially generated data for testing and evaluating edge cases.

4.2 Your Data and Model Training

As described in the Privacy Policy and Terms of Service, we may use your Inputs and Outputs to improve our AI models. Data is de-identified or anonymized before use.

Data from Gmail, Google Calendar, or Google Drive, if you connect a Google account, is never used to train, evaluate, or improve our AI models. See Privacy Policy Section 13.3.

You can opt out by emailing privacy@cfosilvia.com. Opting out does not degrade service quality. It does not affect runtime processing (which is required to deliver the Services) or creation of Anonymized Data.

Regardless of opt-out, data flagged for safety review, submitted as explicit feedback, or reported for policy violations may be used for maintaining safety and integrity.

4.3 Runtime Data

When you use the Services, AI processes your financial data, conversation history, and Inputs in real time to generate Outputs. This runtime processing is distinct from model training and cannot be opted out of while using the Services.

5. Third-Party AI Providers

Silvia uses foundation models and AI infrastructure from third-party providers, listed as service providers in the Privacy Policy. Key facts:

• Third-party providers process your Inputs and deliver Outputs on our behalf.
• We contractually prohibit providers from using your Inputs or Outputs to train their own models, unless separately disclosed.
• Providers are subject to security, confidentiality, and data protection obligations.
• We evaluate providers before engagement and conduct periodic reviews.

Specific providers may change over time as we evaluate performance, cost, and compliance. The current list is maintained in our Privacy Policy.

6. Limitations and Risks

6.1 Accuracy

Outputs may contain factual errors, calculation mistakes, outdated information, or incomplete analysis. LLMs can hallucinate. Transaction categorization is not 100% accurate. Projections are based on assumptions, not certainty.

6.2 Currency

AI models have knowledge cutoff dates. They may not reflect the most recent tax law, regulatory, market, or economic developments. Financial data from linked accounts may be delayed.

6.3 Not Professional Advice

Silvia is not a financial advisor, tax preparer, investment adviser, broker-dealer, or attorney. Outputs are informational tools. Consult qualified professionals for significant decisions. You are solely responsible for your financial decisions.

6.4 Bias

AI models may reflect biases in training data. We work to identify and mitigate bias but cannot guarantee elimination. Report bias concerns to feedback@cfosilvia.com.

6.5 Context Limits

The AI cannot account for information you have not provided and may not fully understand nuances of your personal situation.

6.6 Output Similarity

Other users with similar data or questions may receive similar or identical Outputs.

7. Safeguards and Governance

7.1 AI Governance Framework

• Model Evaluation: Before deployment, models undergo evaluation for accuracy, reliability, safety, and fairness across diverse test scenarios.
• Guardrails: System-level instructions, safety filters, and output validation checks prevent harmful, misleading, or inappropriate content.
• Human Oversight: Ongoing review of output quality, periodic audits, and investigation of user-reported issues.
• Incident Response: Processes to identify, investigate, and remediate AI-related incidents.

7.2 Bias Monitoring

We evaluate outputs across diverse demographics and financial profiles, monitor for disparate impact, investigate user-reported bias, review training data composition, and apply fairness-aware techniques where feasible.

7.3 Security

AI systems are subject to all security measures described in the Privacy Policy, plus protection against prompt injection and adversarial attacks, monitoring of inputs and outputs for anomalies, access controls on model configurations, and logging of model versions and changes.

7.4 Data Minimization

AI is provided only the data reasonably necessary for the requested Output. We do not feed your entire financial history into every request.

8. Your Choices and Controls

9. Anonymized Data and AI

As described in the Privacy Policy and Terms of Service, we create Anonymized Data from user data and may commercialize it. Anonymized Data never includes data from Gmail, Google Calendar, or Google Drive (see Privacy Policy Section 13.3). AI plays a role in both creation and use:

• Creation: AI and algorithmic techniques (generalization, perturbation, differential privacy) transform personal data into Anonymized Data.
• Training: Anonymized Data is used to train and improve our AI models and may be licensed to third parties for their model training, analytics, or research.
• Products: Anonymized Data may be incorporated into commercial data products such as benchmarking reports, market research, and statistical analyses.

Anonymization standards and protections are described in the Privacy Policy (Section 13.1).

10. Regulatory Alignment

This Statement is designed to align with the following frameworks:

10.1 High-Risk Classification Analysis

Under the EU AI Act Annex III, AI systems used for creditworthiness assessment, credit scoring, and certain financial services decisions are classified as high-risk. Silvia’s current AI use cases are informational and educational. Silvia does not make credit decisions, approve or deny financial products, or produce Outputs with legal or similarly significant effects. Accordingly, Silvia’s current AI features are not classified as high-risk under Annex III. We monitor regulatory developments and will update this analysis if we introduce features that cross the high-risk threshold.

11. Accountability and Feedback

11.1 Internal Accountability

• Executive Oversight: Senior leadership sets AI strategy, approves governance policies, and reviews risk assessments.
• Product and Engineering: Responsible for model selection, evaluation, deployment, monitoring, and incident response.
• Privacy and Legal: Ensures AI practices comply with applicable data protection laws and this Statement.
• Trust and Safety: Monitors AI outputs for safety, bias, and quality; investigates user-reported issues.

11.2 User Feedback

We actively seek feedback on our AI systems. Feedback channels include thumbs-up/down on individual responses, manual corrections to categorizations, email to feedback@cfosilvia.com, and customer support at support@cfosilvia.com.

11.3 Continuous Improvement

We will update our practices and this Statement as AI technology evolves, regulations mature, and we learn from feedback. Material changes will be communicated through the mechanisms described in the Terms of Service.

12. Glossary

13. Contact Us

Silvia, Inc.

AI and Transparency Inquiries: feedback@cfosilvia.com

Privacy: privacy@cfosilvia.com

Security: security@cfosilvia.com

General Support: support@cfosilvia.com