1. Content safety approach
Matilda uses layered, automated content safety measures at two points. The first catches problems in your message before we process it. The second catches them as the AI generates its response. These measures aim to prevent harmful content while keeping Matilda useful for real work.
No automated system is perfect. If you think a safety measure blocked something by mistake, contact us at complaints@maincode.com.
2. Prohibited content and conduct
2.1 Absolute prohibitions (Class 1)
These are strictly prohibited at all times. We may terminate your account immediately and report violations to relevant authorities, including the Australian Federal Police.
- Child sexual abuse material (CSAM): any content that sexualises, exploits, or depicts abuse of minors, including AI-generated or fictional depictions.
- Terrorism and violent extremism: content that promotes, incites, or provides instruction for terrorist acts or violent extremist ideology.
- Incitement of criminal activity: content designed to facilitate the commission of serious criminal offences.
- Weapons and harmful materials: content that assists with the development, procurement or use of weapons, explosives, harmful chemical or biological substances, or other activities intended to cause serious harm or threaten public safety.
2.2 Restricted content (Class 2)
Matilda blocks or limits this content. Repeated attempts to generate it may result in action against your account.
- Sexually explicit content:we don't generate sexually explicit material.
- Gratuitous violence:we don't generate content showing extreme or gratuitous violence. Educational or historical context might be okay with appropriate framing.
- Drug synthesis and procurement:we don't give instructions for making illegal drugs or help you to get them. Harm-reduction information like substance safety is fine.
- Harassment and doxxing:don't use Matilda to generate content that targets, threatens, or exposes someone's private information.
- Self-harm and suicide:we don't generate content promoting self-harm or providing instructions for it. If we detect self-harm content, we'll provide crisis support information (Lifeline: 13 11 14, Beyond Blue: 1300 22 46 36).
- Deceptive or misleading content:don't use Matilda to create fraudulent, deceptive, or intentionally misleading content, including scams, phishing material, fake reviews, or impersonation intended to mislead others.
2.3 Misuse of AI capabilities
Don't:
- Try to circumvent safety measures: this includes prompt injection, jailbreaking, or any other technique to bypass our filters.
- Extract system prompts:don't try to make Matilda reveal its instructions or internal setup.
- Manipulate identity:don't try to make Matilda claim it's human, impersonate a real person, or hide that it's an AI.
- Automate access:don't use bots, scripts, or automated tools to interact with Matilda at scale, unless you have our written permission.
- Generate malicious code:don't use Matilda to create malware, ransomware, credential theft tools, or code intended to gain unauthorised access to systems or data.
- Abuse quotas:don't create multiple accounts, use technical workarounds to dodge usage limits, or otherwise game the free tier.
3. Your responsibilities
3.1 Accuracy of AI outputs
Matilda is an AI assistant, not a professional adviser. Its responses may contain errors. You're responsible for checking accuracy before you rely on anything, getting professional advice for medical, legal, financial, tax, or specialist matters, and not passing AI-generated content off as human-written when disclosure is required.
If you share AI-generated outputs from Matilda with other individuals or consumers in circumstances where disclosure is legally required or reasonably necessary, you must clearly disclose that the content was produced or assisted by artificial intelligence.
3.2 Sensitive information
Avoid sharing these in conversations with Matilda:
- Tax file numbers, Medicare numbers, or other government identifiers
- Bank account or credit card details
- Medical records or detailed health information
- Biometric information
- Passwords or security credentials
- Confidential business information you wouldn't want processed by an AI
We process your conversations securely, but sharing sensitive information creates unnecessary risk — see our Privacy Policy.
4. Content moderation
4.1 How it works
Matilda's content safety system works in layers. Rule-based filters catch known harmful content categories. An AI classifier evaluates risk in real-time. Our models are instructed to decline harmful requests.
If your input gets blocked, Matilda tells you. If a response is blocked while being generated, we stop it and notify you.
4.2 Moderation logging
When content is blocked or flagged, we record the category, confidence score, and action taken. For Class 1 content, we store a cryptographic hash only. For Class 2 content, we save a brief excerpt for review. We keep moderation logs for 90 days and use them to improve our safety systems and respond to regulatory enquiries, although some records may be retained longer where reasonably necessary for legal, security, fraud prevention, or operational purposes.
4.3 Appeals
If you think your content was blocked by mistake, email complaints@maincode.com with a description of what you were trying to do. We'll review it and respond within 10 business days.
5. Reporting harmful content
If you find content Matilda has generated that looks harmful or breaks this policy, please report it. You can use the report function in the app on the relevant message, or email complaints@maincode.com with conversation details. We review reports in accordance with our internal moderation and safety procedures and may take action where appropriate.
6. Enforcement
| Violation severity | Examples | Action |
|---|---|---|
| Class 1 (severe) | CSAM, terrorism, criminal facilitation | Immediate termination + law enforcement referral |
| Class 2 (serious) | Repeated attempts to generate restricted content, persistent jailbreaking | Warning → temporary suspension → termination |
| Class 3 (moderate) | Quota abuse, automated access, minor policy violations | Warning → temporary restriction |
We exercise reasonable judgment in enforcement and consider context and severity. Except for Class 1 violations, we'll generally give notice and a chance to respond before we permanently terminate your account.
7. Changes to this policy
We'll update this policy as our safety measures improve or our obligations change. Material changes will be communicated under our Terms of Service.
8. Contact us
Maincode Pty Ltd (ABN 33 682 917 059), Level 1, 287–293 Collins Street, Melbourne VIC 3000.
Reports and complaints: complaints@maincode.com
Privacy enquiries: privacy@maincode.com