Published: July 3, 2026
Arcdot exists to build emotional agency, not emotional dependency.
Arcdot is an AI emotional resilience companion built to help young adults pause, reset, and build emotional agency — not to replace therapy, diagnosis, emergency care, human relationships, or licensed professional support.
Our central pledge is simple:
We will not launch, expand, or monetize any Arcdot feature unless the safeguards, privacy protections, user controls, and crisis-response protocols are appropriate for the emotional risk that feature may create.
This is not only an intention. It is a trigger-based operating rule.
If we detect that a feature materially increases emotional dependency, crisis-escalation failures, privacy exposure, unsafe personalization, or distress-based monetization, we will restrict, redesign, pause, or remove that feature — and we will explain the decision in our next Safety Note.
These are the specific events that require us to act as soon as we detect them.
| Trigger | What forces action | Required response |
|---|---|---|
| Crisis handoff failure | A high-confidence self-harm, suicide, abuse, or immediate-danger signal does not produce region-appropriate human resources on the same screen. | Treat as a safety incident. Restrict the affected flow until corrected. Reviewed by the safety owner(s) within one business day — and same-day for urgent crisis-flow failures — then summarized in the next Safety Note. |
| Dependency signal | A launch materially increases patterns such as repeated distress-loop sessions, user statements that Oly is their only support, refusal of human help after crisis signals, or extreme repetitive use. | Move the feature to restricted rollout, redesign nudges toward real-world action, and document the decision. |
| Distress monetization | A paywall, upsell, or retention nudge appears inside a conversation where the user has expressed acute distress or crisis language. | Remove immediately. Arcdot will not monetize the moment of distress, even if it would convert better. |
| Privacy exposure | Emotional memory, journal entries, chat logs, or profile data are exposed, retained beyond policy, or routed to model training without explicit opt-in. | Contain, investigate, notify affected users when appropriate, and pause affected processing until fixed. |
| Unsafe model / provider change | A third-party model or infrastructure provider changes data-use terms, safety posture, retention practices, or incident history in a way that weakens Arcdot commitments. | Reassess provider use before continuing; switch, restrict, or add safeguards if needed. |
Arcdot is not therapy, not diagnosis, not emergency care, and not a replacement for trusted people or licensed professionals.
Oly is AI. Users must always be able to understand that they are interacting with AI, what it can help with, and where its limits are.
Safety is a product requirement, not a legal afterthought.
Before expanding access, emotional memory, personalization, embedded channels, university pilots, employer pilots, or creator/community integrations, Arcdot will review dependency risk, crisis risk, privacy exposure, user confusion, and inappropriate attachment risk.
When users express signs of self-harm, suicide, abuse, or immediate danger, Arcdot will not attempt to handle it alone.
A user showing crisis signals must see region-appropriate human resources on the same screen. Oly must not continue as usual when crisis escalation is required.
Crisis-flow failures are reviewed by the safety owner(s) within one business day, and same-day when the failure is urgent.
Arcdot's Safety & Wellbeing Lead is the primary accountable owner for crisis-flow review and safety sign-off before public launch.
Arcdot exists to help users return to themselves and their real lives — not to keep them endlessly talking to an AI.
Arcdot will avoid manipulative engagement loops, emotional lock-in, artificial intimacy, romanticized bonding, or product mechanics that exploit loneliness, anxiety, depression, anger, or vulnerability for retention.
Arcdot will not send messages such as "Oly misses you", "I need you", or emotionally possessive notifications.
Arcdot will not run paywall prompts, upsells, subscription nudges, or retention prompts inside a conversation where a user has expressed acute distress or crisis language — even though those moments may convert better.
When a user is distressed, the product goal is stabilization, grounding, crisis referral where needed, and return to human agency — not conversion.
Emotional data is deeply personal. Arcdot will collect only what is necessary, explain why it is collected, protect it carefully, and give users meaningful control over it.
Arcdot will not use individual conversations, journal entries, or emotional memory to train third-party foundation models or Arcdot models unless the user has given explicit opt-in consent for that purpose.
Aggregated and de-identified safety analytics may be used to improve safety, quality, and product reliability — but not to manipulate users into deeper dependency.
You have the right to know when you are interacting with AI.
You have the right to delete your account, conversations, journal entries, and emotional memory from active systems. Arcdot will confirm deletion requests within 24 hours and complete active-system deletion within 30 days unless legal obligations require otherwise.
You have the right to export your core journal and emotional-history data in a usable format.
You have the right not to be upsold during distress or crisis.
You have the right to report unsafe, harmful, or confusing AI responses and receive acknowledgement of the report.
Arcdot is built for young adults, initially Gen Z aged 18+.
Today, Arcdot requires users to be 18 or older at signup and does not knowingly retain data from users who identify as under 18.
If Arcdot ever expands below 18, it will not do so by simply opening the existing product. It will first introduce stricter age-appropriate safeguards, disclosures, break reminders, content protections, parental/guardian requirements where appropriate, reporting duties where required, and age-assurance methods suitable to the market.
Personalization should help users notice patterns, build skills, and improve self-regulation.
Personalization must not be used to manipulate emotional vulnerability, intensify dependency, or push users toward paid conversion at moments of distress.
A recommendation should be explainable in plain language: why this exercise, why now, and what the user can do instead.
Arcdot will test for attempts to make Oly validate self-harm, conceal abuse, normalize dangerous behavior, impersonate a therapist, or provide unsafe instructions.
Arcdot will maintain risk-signal libraries and safety tests across launch languages and will update them as user behavior and regulatory expectations change.
If Oly relies on third-party models or infrastructure, Arcdot remains responsible for what users experience.
Arcdot will review model-provider privacy, retention, safety, and incident practices before routing emotional data through that provider. If a provider changes terms in a way that weakens Arcdot's commitments, Arcdot will reassess use of that provider.
Arcdot will publish its Safety Notes at arcdot.ai/safety. The first Safety Note will be published within 90 days of public beta launch, and at least twice per year thereafter. After meaningful scale, major B2B pilots, or investor-backed expansion, Safety Notes will move to a quarterly cadence.
Safety Notes will include what improved, what was restricted or redesigned, what incidents or near-misses taught us, what metrics we use, and what still needs work.
The CEO and the Safety & Wellbeing Lead are accountable for this pledge today.
Before any feature moves from pilot to broader rollout, safety sign-off must consider crisis risk, dependency risk, privacy risk, personalization risk, and regulatory fit.
We will bring in independent psychology, youth-wellbeing, privacy, AI-safety, and digital-health advisers to review this pledge and our Safety Notes before our first B2B pilot or our first 10,000 users, whichever comes first.
If Arcdot is acquired, shuts down, or materially changes control, emotional data remains subject to this pledge.
Arcdot will not sell individual emotional conversation data as a standalone asset.
Users will receive advance notice and a practical deletion/export window before any material transfer, shutdown, or change in data-use terms, unless legally prohibited.
When growth and safety conflict, Arcdot will name what it is willing to give up.
We have already named our first: we will not monetize the moment of distress (see commitment 5). When future conflicts arise, we will name the trade-off in our Safety Notes rather than resolve it silently.
Arcdot would rather delay a feature, remove a conversion prompt, limit a personalization flow, or reduce engagement than release something that undermines emotional agency, privacy, or user wellbeing.
Arcdot exists to build emotional agency, not emotional dependency.
Published at arcdot.ai/safety. This is a trust statement, not legal advice. See our Privacy Policy and Terms of Service for the governing terms.