Ethical AI Use

1. Telling you it is an AI

Everything Aelius says, writes, speaks or draws is produced by an AI system, and we tell you so before your first chat. No person writes or reviews replies as they are made, and the voice in voice mode is synthetic. The assistant is instructed to say plainly that it is an AI whenever someone asks, including in styles designed to read more naturally. Images it generates carry a machine-readable marker identifying them as AI-generated (the IPTC digital source type, in the file's metadata). The details are in §5.6 of our Terms of Service.

2. What we aim for

• Useful to adults. Answer the question that was asked, including difficult ones, rather than refusing on a keyword.
• Honest about limits. Say what the model gets wrong, and never claim more safety or accuracy than exists.
• Private by default. Collect what the service needs, never train on your conversations, never sell data, and let you delete it.
• Hard lines where harm to others is clear. Minors, non-consensual imagery of real people, targeted harassment and attacks on others are refused and enforced.

3. What we actually do

• Image safeguards. Every image request is screened before it is generated, and every finished image is checked again before it is saved or shown; content involving minors is blocked and handled under our CSAM policy.
• Suicide and self-harm protocol. Messages that read as the sender's own thoughts of suicide or self-harm are flagged to the model, which is instructed to respond with care and refer the person to a crisis line (988 in the US, Samaritans 116 123 in the UK and Ireland, findahelpline.com elsewhere). The protocol is published in §6.2 of the Terms.
• Manipulation attempts. Messages shaped like attempts to override the model's rules are detected and counted, and the model is made to reason more carefully on them.
• Monitoring. We review refusals, blocked requests and reported problems, and change the system when it gets things wrong in either direction.
• Privacy. The models run on our own hardware rather than a third-party AI service; incognito chats are not saved to your history; see the Privacy Policy for exactly what is stored and for how long.
• Energy. Our GPUs run under power caps below their maximum, and replies are routed to a smaller model when it is enough for the job.

4. Limits we know about

The models we run were trained by others, on data we did not choose and cannot audit, so they carry that data's gaps and biases. We have not run a formal bias audit. The model can be confidently wrong, can misread its own web searches, and can answer differently to the same question. Automated safeguards miss some things and block some things they should not. We would rather tell you this than imply otherwise.

5. Who is accountable

The people who run Aelius are responsible for how it behaves. There is no separate ethics board. If the assistant did something harmful, unfair or wrong, or you want to report a safety problem, write to [email protected] or use the contact page; reports about minors are handled first.

6. Our Approach to Content Safeguards

In short: Aelius AI applies content safeguards, and we design them to be permissive. This section explains what that means in practice and what you should understand before using the service.

6.1 Permissive by Design

Aelius AI is built to engage with a wider range of subject matter than most commercial assistants. Where other products refuse on the basis of a keyword, a tone, or a topic that merely sounds sensitive, we aim to answer the question that was actually asked, including questions touching on mature, controversial, or difficult material.

This is a deliberate design choice rooted in the principle of user autonomy. We believe adult users have the right to access information and tools without paternalistic gatekeeping. It is not an absence of standards: the model is designed to decline requests that fall within our Usage Policy's prohibited categories, and we enforce that policy.

6.2 What This Does Not Mean

A model that answers is not the same as a model that is right. Openness says nothing about accuracy, and safeguards are not a guarantee of safety. Users should understand:

• The model cannot distinguish truth from fiction, it generates statistically plausible text, not verified facts.
• The model's outputs do not represent the views of Aelius AI or its operators.
• A confident, unhesitating response is not necessarily a correct, safe, or advisable one.
• No filtering system is perfect. Safeguards can miss things, and they can also refuse things they shouldn't.

6.3 Your Responsibility as a User

Automated safeguards cannot substitute for your own judgment. You should:

• Never follow AI medical advice without consulting a licensed physician.
• Never follow AI legal advice without consulting a qualified attorney.
• Never follow AI financial advice without consulting a regulated financial advisor.
• Never follow safety-critical instructions generated by AI without expert verification.
• Verify factual claims independently before acting on them.
• Use your own judgment, a system that answers freely relies on you to ask responsibly.

6.4 Our Commitment

We prohibit uses that cause clear harm to others, particularly content involving minors, targeted harassment, and malicious use against third parties, and we enforce those prohibitions technically and contractually. We are transparent with users about how the model behaves, require explicit agreement to our Terms of Service before use, and describe our safeguards honestly rather than overstating them.

Ethical AI, in our view, includes being honest about what our AI is and what it is not, rather than creating a false sense of safety through hidden filtering that users are never told about.

7. Conclusion

Ethical AI use is not a destination but a journey. At Aelius AI, we are committed to transparency, including transparency about how our safeguards work and the limitations of all AI systems. We recognize that as AI technology evolves, so too must our ethical framework. We invite ongoing dialogue with our users, the broader AI community, and society at large as we work towards a future where AI benefits all of humanity.

The most ethical thing we can do is be honest: this AI can get things wrong, sometimes badly wrong, and you should never blindly trust it. Use it as a tool, not an authority.