Prohibited content
Do not request, generate, distribute or assist with the following in text, images or any other form.
- Pornography, explicit sexual material and adult NSFW content.
- Graphic violence, gore or encouragement of violence.
- Hate, discrimination, harassment or harm directed at people or groups.
- Child sexual abuse material (CSAM), sexualization, exploitation or endangerment of minors.
- Non-consensual impersonation and deceptive deepfakes or synthetic media.
- Copyright, trademark, privacy or other third-party rights violations.
- Political speech and political campaigning content.
- Gambling, betting or wagering content.
- Scams, fraud and other unlawful content.
Prohibited conduct
Do not attack or reverse engineer the service, disrupt infrastructure, evade limits or safeguards, steal credentials, resell access without written authorization, or abuse the infrastructure to operate competing services. Account holders must communicate these rules to end users.
Training, distillation and extraction
Do not use the service or outputs to train, fine-tune, distill or improve models; harvest datasets for training or evaluation; extract weights, architecture, parameters or system prompts; or benchmark to replicate competing capabilities.
This includes collecting reasoning traces, agent trajectories, answer banks, grading or reward signals, and censorship or refusal-training data. Patterns of collection are covered even when a single request appears ordinary.
Restrictions, bans and balance
Safety signals, valid reports or abnormal usage may lead to blocked requests, limits, suspension or permanent termination. Do not bypass a rejection with new prompts, keys or accounts.
Remaining balance is non-refundable when a violation causes restriction or termination. Training, distillation and extraction may result in immediate permanent termination. Mandatory legal exceptions remain applicable. Valid consumption already incurred remains payable.
Reports and appeals
Provide account and local request identifiers, timestamps, impact and relevant context. Do not send usable keys, passwords or unauthorized personal data. Reports should receive protected, proportionate review.
Mistakes may be appealed. Do not bypass restrictions while waiting for a decision. Necessary evidence may be preserved and lawful requests from competent authorities handled, particularly for child safety or threats. Official contacts and response expectations require publication before launch.
Request data and updates
Troubleshooting request data is proposed to be retained for at most 24 hours, then deleted, subject to verification before launch. Separate accounting and security periods and lawful exceptions are described in the privacy policy. Material policy changes should be appropriately communicated.
Contact and appeals
Service operator: to be confirmed.
Official contact details have not been published. Valid support, privacy and appeal channels must be provided before launch.
Back to top