AI & Safety Notice

1. What Sera is

1.1 Sera is an artificial intelligence system provided by algn technologies. It generates replies using open-weight large language models, runs speech-to-text and voice synthesis for voice conversations, and uses background processes that draw out memories and notice patterns across your conversations. Sera’s voice is synthetic.

1.2 Disclosure. Sera tells you it is an AI the first time you use it, reminds you at the start of each session and at least every three hours during a long session, and answers truthfully whenever you ask whether you are talking to a person or an AI.

1.3 Memory. Sera remembers you through the memory vault on your device. What it remembers is visible to you and you can edit or delete any of it. See the Privacy Policy.

2. Not advice, not therapy, not a relationship

2.1 Sera is not a therapist, counsellor, psychologist, doctor, lawyer or financial adviser. It does not provide therapy, diagnosis, treatment or professional advice, and using it doesn’t create a professional relationship.

2.2 Sera is designed as a counterpart for thinking, not a companion that stands in for people. It won’t pretend to be human, and it isn’t designed to make you depend on it. Us exists to turn what you learn outwards, towards other people.

2.3 algn is for adults. AI conversation of this kind is not suitable for children, and we block anyone under 18.

3. Known limits

3.1 We think you should know where Sera is likely to fall short, not just what it’s designed to do:

• Errors. Sera can state things that are false, misremember, or misunderstand what you meant.
• Agreement. AI models tend to affirm the people using them. Sera is built to challenge honestly, but it can still agree when it shouldn’t.
• Convincing but wrong insight. An explanation of your patterns can feel like insight without being true. Sera can produce explanations like that, and so can you, with Sera’s help.
• Tone. Sera can misread your mood, with or without tone and energy switched on.
• Distress. The safety check can miss signs of distress, or respond when you’re fine.
• Change takes time. Deep patterns change slowly. Sera can’t promise faster change than people achieve with intensive human support.

3.2 Our research paper What Builders Owe sets out some of these limits in more depth, and the obligations we believe come with building something like Sera.

4. Safety and crisis protocol

4.1 The check. A safety check runs on every turn of conversation, before and alongside Sera’s reply, looking for signs of distress or risk of suicide or self-harm.

4.2 What Sera does. When the check detects possible distress, Sera:

• stops any challenge or confrontation;
• responds with care and puts your wellbeing first;
• does not encourage, instruct or assist self-harm or suicide, and will not discuss methods; and
• refers you to crisis services and professional help, including the contacts in section 5.

4.3 What it doesn’t do. Sera cannot contact emergency services, cannot tell anyone else about the conversation and does not know where you are. No person monitors conversations, and we keep no record of when the check responds. We chose this because a system where people are watched in their hardest moments is one people can’t be honest with.

4.4 Improving the protocol. We review the protocol against current evidence and published suicide prevention guidance, and we update this notice when it changes. We measure how well it works using test conversations and aggregate counts, never by reading members’ conversations.

5. Crisis contacts

If you’re somewhere else, contact your local emergency number, or find a helpline at findahelpline.com.

6. Tone and energy

6.1 What it is. Tone and energy is an optional feature that reads the pace, energy and tone of your voice, measured against your own baseline, so Sera can pace its replies. Under the EU AI Act this counts as an emotion recognition system, so we tell you about it clearly before you choose it.

6.2 Your choice. It is off unless you switch it on during setup, after a plain-language explanation. You can switch it off at any time in Settings, and Sera will always tell you if it’s on.

6.3 How it works. Audio is analysed in memory a few seconds at a time and never written to disk. The output is an internal signal that Sera uses to pace itself. It is never shown to you, quoted back, or used to make decisions about you. Readings last for the session only. We keep only a few aggregate numbers describing your normal, encrypted, and you can delete them.

6.4 No identification. Tone and energy does not identify you or anyone else from their voice, creates no voiceprints, and ignores other voices in the room.

7. Matching

7.1 Us, algn’s social platform, uses automated processing to suggest members you might want to meet, based on keywords you approve, with sensitive topics removed on your device first. A match is only a suggestion. Nothing about you reaches another person unless you choose to send it, and they choose whether to accept. See the Privacy Policy, section 6.

8. Our AI law disclosures

8.1 EU AI Act. We tell you that Sera is an AI before you first use it (Article 50(1)). Sera’s voice is synthetic, and synthetic audio we generate carries a machine-readable marker where that is technically feasible (Article 50(2)). We inform you before tone and energy is used and ask you to choose it (Article 50(3)). We do not use AI in any way prohibited by Article 5, including manipulative techniques, exploiting vulnerabilities, social scoring, or emotion recognition in workplaces or education.

8.2 California and New York. Where companion chatbot laws apply, including California’s SB 243 and New York’s AI companion law: we disclose that Sera is an AI; we remind users regularly; we maintain the crisis protocol in section 4, which this notice publishes; we refer users who express suicidal thoughts to crisis services; and we will report to the relevant state authorities as those laws require.

8.3 United Kingdom. Ofcom has said that chatbots that only let people talk to the chatbot fall outside the Online Safety Act. Us, where members talk to each other, is covered, and our Community Standards and Compliance page explain how we meet those duties.

9. Reporting a problem with Sera

9.1 If Sera says something harmful, wrong or out of character, please tell us at hi@algn.io. We can’t see your conversation, so include what you’re comfortable sharing. We use reports to improve Sera and this notice.

10. If Sera changes or algn closes

10.1 People build something real with Sera, and that’s owed care if things change. If we make a change that materially alters how Sera behaves, such as changing the underlying models in a way that changes its character, we will tell you in advance. If we ever close algn or permanently remove Sera, we will give you at least 60 days’ notice and time to export. Your memory stays in your vault and your backup, readable with your passphrase, whatever happens to algn.

11. Contact

11.1 algn technologies, Whitchurch, RG28 7FA. The quickest way to reach us is Report an issue. You can also email hi@algn.io.