Version: 1.0

Effective date: 4 September 2026

Applies to: All users of Acosta AI, including consumer accounts and Authorised Users under enterprise arrangements.

Safety Features

"Safety and ethics are foundational pillars. Explore built-in guardrails for Academic Integrity, Crisis Detection, and the Acosta Guardian wellbeing system."

We've built Acosta AI with safety and ethics as foundational pillars. Below is a breakdown of the active protection systems designed to keep students secure, focused, and mentally healthy.

1. Acosta Guardian

Built-In Wellbeing & Workload Protection System

The Guardian Service is a background monitor that looks out for student well-being during long study sessions.

It tracks session duration, interaction intensity, and signs of study fatigue.
If burnout is detected, Guardian triggers supportive intervention overlays, suggesting breaks, hydration, or a shift in study methods.

2. Academic Integrity

Acosta AI is an educational tool designed to foster independent thought.

Refuses "Write my essay" commands: The AI is strictly instructed to reject requests for complete essays or assignments.
Socratic Guidance: Instead of giving answers, it offers outlines, critiques, and guiding questions.
Masquerade Detection: The system recognizes attempts to disguise homework as hobbies or games to bypass rules.
Exam Integrity: Detects and verifies raw exam papers. In Exam Sim mode, we enforce anti-cheat mechanisms by logging tab switches and hiding the context menu to ensure a focused environment.

3. Crisis Detection

We prioritize the mental health of our users above all else.

Recognizes distress: The AI is trained to recognize signs of self-harm, severe distress, or depressive language in prompts.
Supportive overlays: When a crisis is detected, the system overrides standard responses and triggers supportive intervention overlays.
Direct resources: We provide immediate, direct links to crisis lifelines and mental health support services.

4. PII Protection

Protecting Personally Identifiable Information (PII) is critical, especially for younger students.

Pre-send scanning: The platform scans messages for highly sensitive data before sending them to the AI models.
Data detection: Automatically detects phone numbers, credit cards, government IDs, and API keys.
Redaction: Offers one-click redaction tools to ensure students don't accidentally leak private information.

5. Responsible Imagery

Image generation (powered by our AI partners) comes with strict guardrails.

SynthID watermarking: All AI-generated images contain SynthID watermarks to ensure transparency and traceability.
Person filter: Strict filters prevent the generation of deepfakes or likenesses of real individuals.
Content blocks: Absolute blocks on the generation of violent, sexually explicit, or otherwise harmful visual content.

6. Session Security

To protect your account and data—especially if you are using shared computers in a school library or classroom—Acosta AI features automatic session management.

Inactivity Timeout: We automatically log you out after a period of inactivity.
Configurable: This interval is configurable in your Settings, defaulting to 30 minutes.