AI Safety Policy

(ü)ALTER is an AI Agent platform that lets independent professionals (coaches, consultants, educators, designers, etc.) create an AI digital twin trained on their own methodology, and deploy that twin to interact with their clients on their behalf.

(ü)ALTER is not:

• A licensed provider of medical, legal, or financial advice
• A replacement for human professional judgment — it is a lever to amplify expert work
• An autonomous AI operating independently — you (the expert) are always the final responsible party for its behavior
• A “general-purpose AI” or a ChatGPT substitute

Every (ü)ALTER agent is designed by default to be observable and intervenable by the expert who owns it:

• The expert can observe in real time every conversation their AI agent has with every client
• The expert can intervene at any moment, override the AI's reply, or pause the agent entirely
• When facing situations outside its training or in high-stakes contexts, the AI agent escalates to the expert rather than acting unilaterally
• Clients clearly know they are interacting with an AI (through explicit indicators + the agent disclosing it on introduction)
• Clients can request to speak with a human expert at any moment

(ü)ALTER agents are powered by carefully vetted frontier AI models from providers including Anthropic (Claude) and OpenAI (GPT family), depending on the scenario. The full provider list is in Privacy Policy §5.

Key commitments:

• We do not use your uploaded professional methodology to train any general-purpose AI model — your content is used solely to operate your own agent
• We have zero-retention or 30-day-retention agreements with every AI provider that prohibit them from using your prompts to train their models
• Client conversation content is passed to model providers only for the minimum duration necessary, never for their training

Even when the expert holds relevant credentials, (ü)ALTER agents must not be deployed in the following contexts:

• Medical diagnosis or treatment — an AI agent cannot directly diagnose conditions, prescribe medication, or deliver psychiatric treatment to patients
• Legal adjudication or litigation advice — an AI agent cannot issue legally binding opinions without a licensed attorney's sign-off review
• Financial and investment advice — an AI agent cannot make specific investment-decision recommendations without supervision by a licensed financial advisor
• Crisis intervention — conversations involving suicide, self-harm, domestic violence, or imminent physical safety must immediately escalate to a human; the agent must not handle them alone
• Interactions with minors — (ü)ALTER is for professional users and clients aged 18 and above
• Fraud, harassment, false identity — must not be used to impersonate other real people or companies, or to commit any fraudulent act

Violations will result in immediate account termination (see Terms of Service §10).

Large language models have inherent limitations that you and your clients should be aware of:

• Hallucination: AI can confidently produce inaccurate factual claims. We mitigate with retrieval-augmented generation and retrieval verification, but cannot fully eliminate it.
• Bias: models inherit societal bias from their training data. We continuously monitor and mitigate via system prompts and instruction tuning.
• Context limits: long-term memory in AI is an active research frontier. Important client context may fade across long conversations.
• Adversarial inputs: clients may attempt prompt injection to coax the agent outside its intended behavior. We implement multi-layer defenses, but treat them as defenses, not perfect barriers.

• Pause / shut down the agent: stop your AI agent from receiving new messages with one click
• View full conversations: your dashboard shows the complete conversation history with every client
• Override and rewrite: edit the agent's reply before it sends, or correct it after the fact
• Export your data: export all content in portable format (see Privacy Policy §6)
• Request deletion: request deletion of your content at any time

If you observe (ü)ALTER behaving unsafely, identify abuse risk, see potential for client harm, or suspect platform misuse for fraud or abuse, email support@sensebeing.ai with the subject “Safety Report”. We will triage within 24 hours. For emergencies (self-harm, suicide, threats of violence), also contact your local emergency services immediately.