AI and Machine Learning Disclosure

Frontal Labs, Inc. ("Frontal," "we," "us," or "our") provides access to features and services powered by artificial intelligence and machine learning technologies (collectively, "AI-Powered Features"). This AI and Machine Learning Disclosure ("Disclosure") describes the nature, capabilities, limitations, and risks of these AI-Powered Features. This Disclosure supplements and is incorporated into our Terms of Service.

1. SCOPE OF AI-POWERED FEATURES

Frontal's AI-Powered Features include, as applicable to your subscription:

(a) AI Gateway and LLM Routing. Infrastructure for routing requests to third-party large language models (LLMs) and embedding models from providers including Anthropic, OpenAI, Google, Mistral AI, and others. The gateway provides unified API access, authentication, rate limiting, logging, and cost management across model providers.

(b) Model Hosting and Inference. Managed infrastructure for hosting and serving open-source and custom AI models for inference, including LLMs, embedding models, and specialized models.

(c) AI Agents. Tools and infrastructure for building, deploying, and managing AI agents that can perform multi-step tasks, interact with APIs, use tools, and execute workflows autonomously or semi-autonomously.

(d) Fine-Tuning Services. Infrastructure and tooling for fine-tuning base models on customer-provided datasets.

(e) Retrieval-Augmented Generation (RAG). Infrastructure for combining LLMs with external knowledge retrieval from vector databases, search indices, and connected data sources.

(f) Code Generation and Assistance. AI-powered code generation, completion, review, and analysis features.

(g) Embedding Generation. API access to text embedding models for semantic search, clustering, classification, and other vector-based operations.

This list may be updated as we add, modify, or discontinue features.

2. THIRD-PARTY AI MODELS

2.1 Many AI-Powered Features involve access to AI models developed, trained, hosted, and operated by third-party providers ("Third-Party Models"). These providers include entities listed in our Subprocessors List.

2.2 Each Third-Party Model is subject to the capabilities, limitations, content moderation, and usage policies of its provider. We do not control and are not responsible for:

the content, accuracy, or completeness of outputs generated by Third-Party Models;
the availability, performance, or reliability of Third-Party Models;
changes to Third-Party Model behavior, capabilities, pricing, or availability by the provider;
data handling or privacy practices of Third-Party Model providers (except as set forth in our Data Processing Agreement where applicable);
the safety, alignment, or content filtering implemented by Third-Party Model providers; or
the training data, biases, or limitations of Third-Party Models.

2.3 We implement technical measures, including content filtering, input and output logging, and rate limiting, to help manage risks associated with Third-Party Model usage. These measures are designed to reduce, but cannot eliminate, the risks described in this Disclosure.

3. KNOWN LIMITATIONS OF AI-POWERED FEATURES

3.1 Probabilistic Nature. AI models, including LLMs and embedding models, are inherently probabilistic systems. They generate outputs based on statistical patterns learned from training data. The same input may produce different outputs across different invocations, models, or provider configurations.

3.2 Factual Accuracy. AI-Powered Features may produce outputs that are factually incorrect, incomplete, misleading, or nonsensical ("hallucinations"). You must independently verify any factual claims, calculations, code suggestions, analytical outputs, or other information generated through AI-Powered Features before relying on them.

3.3 Bias and Fairness. AI models may exhibit biases present in their training data, including biases related to race, gender, ethnicity, religion, age, sexual orientation, socioeconomic status, and other characteristics. Outputs may reflect, amplify, or reinforce these biases. We cannot guarantee that AI-Powered Features will produce outputs free from bias.

3.4 Inappropriate Content. Despite content moderation measures, AI-Powered Features may generate content that is offensive, harmful, toxic, inappropriate, or otherwise objectionable. We continuously work to reduce such occurrences but cannot eliminate them.

3.5 Safety and Alignment. AI models, particularly in agentic configurations, may take actions or produce outputs that were not intended or anticipated. Agent behaviors may be unpredictable, especially in edge cases, adversarial inputs, or novel scenarios not represented in training data.

3.6 Context and Recency. AI models have knowledge cutoffs determined by their training data. Models may not have accurate or current information about recent events, developments, or data. Some models may incorporate real-time search or retrieval mechanisms; the currency of such mechanisms depends on the underlying providers and implementations.

3.7 Consistency and Reproducibility. Outputs from AI-Powered Features are not guaranteed to be consistent or reproducible across invocations. Differences in model versions, provider configurations, sampling parameters, and system prompts may affect outputs.

3.8 Language and Localization. AI-Powered Features may perform differently across languages, with generally higher performance in English and widely used languages. Performance in less common languages or dialects may be reduced.

3.9 Domain Expertise. AI-Powered Features are not substitutes for professional judgment in specialized domains, including legal, medical, financial, engineering, or safety-critical applications. You must not rely on AI outputs as professional advice.

3.10 Maximum Context and Token Limits. AI models have finite context windows and token limits. Inputs exceeding these limits may be truncated. Long conversations or documents may exceed model context capacity, resulting in loss of earlier context.

4. YOUR RESPONSIBILITIES

4.1 Human Oversight. You are responsible for maintaining appropriate human oversight of AI-Powered Features, particularly in high-risk or consequential use cases. Human review must be implemented commensurate with the risk profile of your use case. We strongly recommend human review of outputs before they are used in:

decisions with legal, financial, or material consequences;
content intended for publication or public distribution;
code deployed to production environments;
safety-critical or infrastructure-critical applications;
decisions affecting individual rights, benefits, or opportunities; and
high-stakes or irreversible actions taken by AI agents.

4.2 Evaluation and Testing. You are responsible for evaluating and testing the suitability, accuracy, and safety of AI-Powered Features for your specific use case, including testing for biases, errors, edge cases, and failure modes relevant to your application.

4.3 Disclosure to End Users. If you deploy or integrate AI-Powered Features into products or services made available to your end users, you are responsible for providing appropriate disclosures regarding the use of AI, including the limitations and risks described in this Disclosure. You must not misrepresent the capabilities, accuracy, or reliability of AI-generated outputs to your end users.

4.4 Compliance. Your use of AI-Powered Features must comply with all applicable laws and regulations governing AI and automated decision-making, including:

the EU AI Act (Regulation 2024/1689);
applicable data protection, privacy, and consumer protection laws;
laws prohibiting unfair, deceptive, or abusive acts or practices;
laws governing automated decision-making and profiling;
sectoral regulations applicable to your industry or use case; and
the Acceptable Use Policy and all other Frontal policies.

4.5 Intellectual Property. As between you and Frontal, you retain ownership of Customer Data as set forth in the Terms of Service. However, the legal status of intellectual property rights in AI-generated outputs is an evolving area of law that varies across jurisdictions. We make no representation or warranty regarding:

the copyrightability or patentability of AI-generated outputs;
the originality of AI-generated outputs or freedom from third-party intellectual property claims;
the application of fair use, fair dealing, or similar doctrines to AI training and generation; or
any ownership rights you may or may not acquire in AI-generated outputs.

You are responsible for assessing the intellectual property implications of AI-generated content in your specific use case and jurisdiction.

5. SECURITY AND PRIVACY CONSIDERATIONS

5.1 Input Data. Inputs submitted to AI-Powered Features (including prompts, documents, and context) are processed to generate outputs. Our handling of input data is governed by our Privacy Policy, Data Processing Agreement, and Data Protection Policy.

5.2 Third-Party Model Providers. When you use AI-Powered Features that involve Third-Party Models, your inputs and generated outputs are transmitted to the applicable model provider. The data handling practices of model providers are governed by their respective terms and privacy policies. Refer to our Subprocessors List for information about model providers and our Model Provider Reference for a summary of each provider's data handling, SLA, and IP indemnity commitments. Additional data processing terms applicable to specific model providers are set forth in our Data Processing Agreement.

5.3 Prompt Injection and Adversarial Inputs. AI-Powered Features, particularly agent systems and RAG pipelines, may be susceptible to prompt injection, indirect prompt injection, jailbreaking, and other adversarial techniques. You are responsible for implementing appropriate input validation, output filtering, and security measures in your applications.

5.4 Training on Customer Data. Frontal does not use Customer Data to train, fine-tune, improve, or evaluate any artificial intelligence or machine learning models. This commitment applies to all models operated, hosted, served, or otherwise controlled by Frontal.

Frontal maintains technical and organizational controls designed to prevent Customer Data from being incorporated into model training datasets or used for model improvement purposes.

Customer Data may be processed solely to provide, secure, maintain, troubleshoot, monitor, and support the Services.

Frontal only integrates with third-party model providers that contractually agree not to use Customer Data submitted through Frontal for training or improvement of their models. For provider-specific details on data handling, see our Model Provider Reference.

6. MODEL EVALUATION AND PERFORMANCE

6.1 We evaluate Third-Party Models and our own infrastructure using a range of benchmarks and metrics. Results of these evaluations may be available through our Documentation or upon request. Results are specific to the evaluation methodology and configuration used and may not reflect performance in your specific use case.

6.2 We do not guarantee or warrant that any AI-Powered Feature or model will meet specific accuracy, latency, throughput, or quality thresholds except as expressly set forth in an applicable Service Level Agreement.

7. DISCLAIMERS

7.1 AI-POWERED FEATURES ARE PROVIDED ON AN "AS IS" AND "AS AVAILABLE" BASIS. TO THE MAXIMUM EXTENT PERMITTED BY LAW, FRONTAL DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, REGARDING AI-POWERED FEATURES, INCLUDING ANY WARRANTY OF ACCURACY, COMPLETENESS, RELIABILITY, FITNESS FOR A PARTICULAR PURPOSE, MERCHANTABILITY, OR NON-INFRINGEMENT.

7.2 WITHOUT LIMITING THE GENERALITY OF THE FOREGOING, FRONTAL MAKES NO REPRESENTATION OR WARRANTY THAT:

OUTPUTS GENERATED BY AI-POWERED FEATURES ARE ACCURATE, COMPLETE, CURRENT, FIT FOR YOUR PURPOSE, OR FREE FROM ERRORS, BIAS, OR OBJECTIONABLE CONTENT;
AI-POWERED FEATURES WILL BE AVAILABLE WITHOUT INTERRUPTION AT ANY PARTICULAR LEVEL OF PERFORMANCE;
OUTPUTS DO NOT INFRINGE THIRD-PARTY INTELLECTUAL PROPERTY RIGHTS;
THIRD-PARTY MODELS WILL BE AVAILABLE OR WILL NOT BE MODIFIED OR DISCONTINUED BY THEIR PROVIDERS; OR
AI-POWERED FEATURES COMPLY WITH LAWS APPLICABLE TO YOUR SPECIFIC USE CASE.

7.3 THE LIMITATIONS OF LIABILITY SET FORTH IN THE TERMS OF SERVICE APPLY TO YOUR USE OF AI-POWERED FEATURES.

8. CHANGES TO THIS DISCLOSURE

We may update this Disclosure as AI technology, regulations, and our Services evolve. Material changes will be communicated in accordance with our Terms of Service.

9. CONTACT

Frontal Labs, Inc. Attn: AI Governance 131 Continental Drive, STE 305 Newark, DE 19713 United States of America Email: legal@frontal.dev