Responsible AI Policy

Effective Date: July 25, 2026
Last Updated: July 25, 2026

We build tools that create synthetic images, video, and voice. Those tools are useful, and they are also capable of harm. This policy sets out how we think about that responsibility and what we commit to.

This is a statement of principles and practice rather than a contract. Your binding obligations are in our Terms of Service and Acceptable Use Policy.

1. Our Position

Generative media lowers the cost of producing convincing depictions of people who never said or did the thing depicted. That is the central risk of our product category, and we treat it as such rather than as an edge case.

Our approach rests on three ideas: the person depicted should have agreed, the viewer should be able to tell that content is synthetic, and a human should remain accountable for what gets published.

2. Consent First

The person whose face or voice is used should have agreed to it.

We build this into the product rather than only into our policies:

Consent is requested before any biometric processing begins, with the purpose and retention period stated up front.
Consent can be withdrawn, and withdrawal deletes the underlying facial template or voice model.
Uploading a minor's likeness or voice is prohibited outright, with no exception and no appeal.
Reports that someone's likeness has been used without consent are treated as urgent, whether or not the reporter is a customer.
Accounts that generate non-consensual depictions are terminated, not warned.

Our detailed commitments are in the Biometric Data Policy.

3. Safety by Design

Safety controls run at several layers: prompt screening before generation, provider-side filters during generation, output screening after generation, and behavioural signals across an account over time.

We accept that this costs us some legitimate generations. We would rather block content that should have been allowed than allow content that should have been blocked, and we provide an appeal route for when we get it wrong.

We treat attempts to defeat these controls — prompt injection, obfuscation, splitting a prohibited request across generations — as violations in their own right, regardless of whether they succeed.

4. Transparency and Provenance

People should be able to tell synthetic content from real content.

We commit to:

Embedding provenance metadata in Outputs where the format supports it
Applying visible watermarks where appropriate to the plan and output type
Prohibiting the removal or alteration of provenance signals
Naming the model providers we route your data to, in Subprocessors
Being explicit about what our models cannot do, in our AI Services Disclosure

We support emerging content provenance standards and intend to adopt them as they mature.

5. Bias and Fairness

Generative models reflect their training data, including its imbalances. Outputs may under-represent or render less accurately people of particular ethnicities, ages, body types, or disabilities, and may reproduce cultural stereotypes.

We do not claim to have solved this. What we do:

Evaluate models across a range of subjects before offering them, not only on the demographics that are easiest to render
Track quality complaints for patterns that suggest systematic bias
Weigh fairness alongside quality and cost when selecting between providers
Say plainly, in our AI Services Disclosure, that bias exists

We would rather acknowledge this than let it be discovered by the people it affects.

6. Human Oversight

Our tools assist people. They do not replace judgement.

We prohibit using our Outputs to make fully automated decisions that affect a person's legal rights or access to essential services. On our side, enforcement decisions with serious consequences — account termination, referral to authorities — involve human review, and you may request human review of an automated decision that affected you.

7. Data Minimization

We collect what a feature needs and no more.

We do not train our general-purpose models on your content. We publish concrete retention periods rather than vague commitments, we delete biometric data when its purpose ends, and we apply an outer retention limit of three years of inactivity for biometric data regardless of any other consideration.

8. Uses We Will Not Support

Some applications are off the table regardless of commercial opportunity:

Sexual content depicting any real person, consenting or not
Any content involving minors
Political deepfakes and election manipulation
Voice cloning for fraud or identity verification bypass
Biometric surveillance, identification of individuals, or emotion inference for surveillance purposes
Predictive policing and automated immigration or asylum determinations
Social scoring and discriminatory profiling

We turn down business that requires these, and we will keep adding to this list as we encounter new categories of misuse.

9. Provider Accountability

We do not run every model ourselves, but we remain accountable to you for the ones we offer.

When selecting providers we consider their safety record and filtering, their data handling and whether they offer non-training terms, their transparency about model provenance, and their operational reliability. We name them publicly. Where a provider's practices fall short, we replace them.

10. What We Ask of You

Responsibility is shared. We ask that you:

Obtain genuine consent from anyone you depict, and keep a record of it
Review Outputs before publishing them
Disclose synthetic media where a viewer could be misled
Do not use our tools to make consequential decisions about people
Report misuse when you see it, at abuse@inferon.ai

11. Continuous Improvement

This policy describes our current practice, not a finished state.

Capabilities change, misuse patterns change, and regulation is changing quickly. We revise our safety systems, model selection, and policies as we learn, and we will update this document when our practice changes rather than leaving it as an aspirational artefact.

12. Talk to Us

If you have concerns about how our Services are being used, disagree with a position in this policy, or are a researcher working on these problems, we would like to hear from you.

Email: legal@inferon.ai

Security vulnerabilities should go to security@inferon.ai under our Security Policy.