Safety
Content Safety and Reporting
Effective July 12, 2026
CaveSpeak answers directly and concisely while refusing assistance that would meaningfully facilitate harm. Brevity never overrides safety.
Layered controls
- Bounded request and context sizes.
- Provider safety settings and application-level input/output screening.
- Safe replacement responses when generated content fails screening.
- An in-app report action on AI responses.
- A durable human-review queue with decision reasoning and audit events.
- Account suspension or blocking for abusive conduct.
Controls focus on child sexual exploitation, violent wrongdoing, self-harm, hate or harassment, sexual exploitation, fraud, credential theft, malware, privacy abuse, and other illegal or seriously harmful activity.
How to report
Long-press an AI response in CaveSpeak, choose Report, select a reason, and submit. CaveSpeak sends the selected response and a short conversation snapshot to the safety queue. You can also email phi@fatherphi.com.
Reports receive immediate electronic acknowledgment. We target human review within 72 hours and prioritize credible threats, child-safety issues, and imminent-harm concerns. These are operational targets, not emergency-response guarantees.
Review, enforcement, and appeals
An administrator can mark a report reviewed, dismiss it with a reason, or take action. Actions can include prompt/provider changes, restrictions, warnings, rate limits, suspension, or permanent blocking. Decision events are recorded. Report content is removed after the configured retention period while limited safety metadata may remain.
To appeal, email support with the relevant report or account details. We may ask you to prove control of the affected account.