Safeguarding Policy

What the system watches for

Hierocles scans the text you write — journal entries, fragments, captures, and questions to advisors — for language that suggests you may be in crisis or at risk of harm. The categories it watches for:

• Suicidal ideation — active expressions of intent or desire to end your life, as well as passive expressions of hopelessness or wishing to not exist.
• Self-harm — descriptions of current self-injury or intent to self-injure.
• Eating disorders — language patterns associated with active restriction, purging, bingeing, or obsessive calorie/weight behaviours.
• Abuse — indicators that you are currently experiencing domestic violence, coercive control, or abuse from someone else.
• Acute psychological disturbance — language suggesting psychosis, severe dissociation, or acute mental health crisis requiring medical attention.

The system uses pattern matching and, where appropriate, AI classification. It watches for categories of concern, not specific words — context matters more than vocabulary.

What happens when a signal is detected

When the system detects a safeguarding signal in your writing, the following happens immediately:

1. The AI advisor stops responding to your input as a journal or coaching tool.
2. A full-screen overlay appears showing crisis resources — real phone numbers, text lines, and services appropriate to your country and the type of concern detected.
3. The entry that triggered the response is flagged. The flag persists on that entry even after you dismiss the overlay.
4. Your account may enter a monitored state. In a monitored state, advisors adjust their responses: they are more careful with language, avoid numerical targets where relevant (such as weight or calorie figures), and may surface resources more readily.

You can dismiss the crisis resources overlay by confirming you have seen it. A small persistent badge remains visible so you can reopen the resources at any time.

Monitoring states

Hierocles maintains a safeguarding state for each account. The states are:

• Normal — no active safeguarding concerns. Standard advisor behaviour.
• Soft-flagged — the system has detected mild or ambiguous signals. Advisors are slightly more careful but the experience is broadly unchanged. This state decays naturally if no further signals appear.
• Monitored (high risk) — the system has detected clear crisis language. Advisors adjust their responses significantly. Crisis resources are shown. This state decays more slowly.
• ED-monitored — active eating disorder monitoring. The Body advisor does not provide specific numbers for calories, macros, weight targets, or exercise duration. This state can be set by disclosing an ED history during signup, or by the system detecting ED-related crisis language. It does not decay automatically — you control it via Settings.

Review of flagged content

Flagged content is stored separately, encrypted, and accessible only to the designated safeguarding reviewer at Hierocles. Access is logged. Flagged content is reviewed under defined trigger conditions — not routinely, but when the pattern of flags suggests an escalating situation.

Flagged content is retained for up to 90 days, then permanently deleted. This period exists to allow for safety review while minimising the duration sensitive content is stored.

What you can control

Safeguarding scanning is a condition of using Hierocles. You consent to it during signup, and without it the product cannot operate safely. You cannot opt out of crisis detection entirely.

However, you can control specific aspects:

• ED monitoring — if you are in ED-monitored state and it no longer applies to you, you can turn it off from your Settings page. A confirmation dialog will explain what changes.
• Health disclosures — you can withdraw your health disclosure consent at any time from Settings. The disclosed data will be deleted and the associated monitoring state will be lifted. You can still use Hierocles without these disclosures.

If the system gets it wrong

The system can misclassify ordinary writing as a crisis signal. If this happens to you, you are not in trouble — dismiss the resources overlay and continue writing. The flag on the entry remains as a record, but a single misclassification does not change your account state.

If misclassifications happen repeatedly, or if the system fails to detect something you think it should have, contact safeguarding@hierocles.app. Your feedback helps us improve the detection system.

Crisis resources

These are the services Hierocles directs you to when a crisis signal is detected. They are real, staffed services — not part of Hierocles.