AI Usage Policy
Last updated: 14 May 2026 · admin@scribbla.com
Scribbla uses third-party large language models to power drafting, editing, summarisation, citation and image features. This policy explains which models we use, what data leaves the platform, our no-training commitment, the limits of AI output, and what is prohibited.
1. Models we route to
- Anthropic Claude family — primary drafting and long-context editing.
- OpenAI ChatGPT (GPT-5 family) — alternate drafting, structured-output tasks.
- Google Gemini family — research, multimodal and image-related tasks.
Routing is chosen by feature and may change as providers ship better models. Material changes to the provider list are reflected here.
2. What leaves the platform
For each generation we send the model only what it needs: the prompt, the relevant chapter or excerpt, plus any context you explicitly attach (character notes, research snippets). We do not send your whole library, your account email, or other users' work.
3. Training opt-out
We use these providers under their API/enterprise terms, which contractually prohibit using your prompts or outputs to train their models. We do not, ourselves, train models on your content.
4. Accuracy & hallucination
AI output may be inaccurate, biased, internally inconsistent, or inadvertently similar to existing works. It may invent facts, sources or quotations. You are responsible for reviewing every AI-generated draft for accuracy, originality, lawfulness and suitability before publishing or relying on it. AI output is not legal, medical, financial or other professional advice.
5. Disclosure of AI assistance
We require you to be honest with your readers. You must not misrepresent AI-generated material as human-authored journalism, scholarship, expert advice in regulated fields, or court evidence. Books published through Scribbla include a generic AI-assisted disclosure in the metadata.
6. Prohibited AI uses
The full list lives in our Acceptable Use Policy; in summary, you must not use Scribbla AI to:
- generate CSAM, terrorism content, or other illegal material;
- produce sexual deepfakes or non-consensual intimate imagery;
- create undisclosed deepfakes of real people;
- generate malware, phishing, scam templates, or fraud scripts;
- mass-produce spam, SEO sludge or platform-manipulation content;
- provide regulated-profession advice presented as human-authored;
- infringe copyright by deliberately reproducing protected works at scale.
7. Reporting AI safety issues
If a Scribbla AI feature produces dangerous, illegal or seriously misleading output, please report it to admin@scribbla.com with the prompt and a screenshot. We treat AI-safety reports as priority triage.