Ethical AI Policy
Useful AI. Thoughtful choices. Human responsibility.
Draft for review. This policy describes proposed principles and use expectations. It is not a certification that every model is safe, unbiased, or suitable for every purpose.
01 Purpose & scope
LocalAI Tools, operated by HubtersAI LLC, helps people explore and run AI models on their own devices. Our aim is to make these tools useful while respecting human rights, individual choice, and the people affected by their use.
This policy describes expectations for using our apps, model integrations, and optional data-sharing features. Each model’s license, usage restrictions, and documented limitations also apply. Availability in our catalog is not an endorsement of every possible use.
02 Human oversight
AI should support human judgment. You remain responsible for how you interpret and use its outputs.
- Check important facts, calculations, translations, and generated media before relying on or distributing them.
- Do not use model output as the sole basis for consequential decisions about employment, housing, credit, education, healthcare, legal rights, or access to essential services.
- Use qualified review, appropriate validation, and a meaningful way to challenge decisions when people may be materially affected.
These tools are not a substitute for qualified medical, legal, financial, or other professional advice.
03 Responsible use
Use AI lawfully and with respect for other people. Do not use LocalAI Tools to:
- Facilitate violence, exploitation, abuse, fraud, or other unlawful activity.
- Create or distribute sexual exploitation material involving children or non-consensual intimate content.
- Impersonate people deceptively, clone a person’s voice without appropriate permission, or present fabricated evidence as authentic.
- Harass, threaten, target, or unlawfully discriminate against individuals or groups.
- Access private systems, obtain credentials, or disclose personal information without authorization.
- Violate applicable intellectual property rights, confidentiality obligations, or a model’s license.
Have the rights and permissions needed for the content you submit, including photographs, recordings, voices, and documents involving other people.
04 Fairness & inclusion
Models can reproduce bias, stereotypes, and unequal error rates. Results may vary across languages, accents, demographic groups, disabilities, recording conditions, and other contexts.
Do not treat a model score or label as an objective judgment of someone’s character, identity, ability, or worth. Evaluate performance in the intended context, examine who may be disadvantaged, and provide human review when mistakes could harm people.
We aim to communicate known limitations and improve accessibility. A successful sample or benchmark does not establish fairness across every population or use case.
05 Transparency & evidence
People should understand when AI materially contributes to content or a decision, particularly where authenticity or trust matters. Disclose AI generation or alteration when its omission could mislead someone.
We aim to identify original model providers, licenses, supported configurations, and measured limitations. Speed measurements and result-quality evaluations answer different questions; neither guarantees that a model will perform the same way on every device or input.
We distinguish measured support from unverified capability. We do not treat a model’s presence in the catalog as proof of universal accuracy, safety, or suitability.
06 Privacy & consent
Local inference reduces the need to send model inputs to a server, but it does not make every product action offline. Account features, model downloads, service analytics, support submissions, and optional run sharing may involve network requests.
Optional sharing of run inputs, files, and outputs requires a separate choice. This Ethical AI Policy does not grant permission to collect or share that content. See our Data Sharing Policy draft for the proposed scope and recipients.
Share only information you are entitled to contribute. Your consent does not replace permission required from other people, and accepting a policy does not waive applicable privacy rights.
07 Safety & reliability
Models may hallucinate, misclassify content, miss important details, or generate unsafe results. Some local models may have limited safeguards. Running a model locally does not remove these risks.
Review model-specific requirements before use. Start with a sample, assess outputs against the real task, and stop using a model when its behavior is unreliable or inappropriate for that setting.
We aim to investigate reported issues and correct misleading capability claims. We cannot guarantee error-free outputs or monitor every run performed locally on your device.
08 Accountability & reporting
If you encounter harmful behavior, a misleading model description, a privacy concern, or a reproducible failure, contact us or use the model page’s reporting feature.
Include the model, configuration, and a description of the issue where possible. Avoid sending sensitive examples unless necessary and authorized. Reporting an issue does not authorize unrelated training use of your attachments.
Depending on the issue and the features we control, responses may include correcting documentation, changing an integration, limiting access to a hosted feature, or removing a model listing. We do not promise remote control over models or files already stored on your device.
Material changes to this policy will be reflected in its date and content. New data-sharing purposes require their own appropriate notice and consent rather than relying on this policy.