DEFINITIONS
Creative work. Scripts, films, animations, novels, etc.
Semi-creative work. Notes, summaries, outlines, bibles, pitch decks, brainstorming, etc.
Office work. Emails, schedules, budgets, etc.
BACKGROUND
AI is the latest extension of Big Tech’s larger ambitions for surveillance capitalism. It’s in their financial interest to harvest as much data as possible, including personal or intimate data, and certainly including valuable creative work and intellectual property.
Meanwhile, AI is extremely useful for content-generation tasks. Meeting notes can be rapidly summarised. Generic emails are composed instantly. Documents can be proofread and punched-up with time-liberating speed. Yet these models only exist because they were already trained on everyone’s data.
From now on, we should be careful and not allow valuable creative IP to be harvested without some kind of compensation. Simultaneously, we can exploit these models for non-creative work, in dozens of ways, and become more creative.
AI AND IP
Currently (June 2025), OpenAI doesn’t train its models on any API data (this includes documents uploaded to conversations with ChatGPT). This may change in the future. So it is worth taking the extra precaution of deleting the conversation that includes uploaded documents, once you don’t need it anymore. Deleting the conversation does actually delete that information from OpenAI (as of June 2025). OpenAI does, however, train on files held on Microsoft’s OneDrive.
Anthropic have a clear policy of not training their Claude models on any user inputs.
It is likely but unconfirmed that Google trains its Gemini models on existing Google Docs, Gmail, and other Workspace files. Avoid using Gemini. Consider migrating creative work away from Google cloud services (DropBox or iCloud are safer).
Meta trains its models on Facebook and WhatsApp data, including uploaded files.
RECOMMENDED USAGE
For storage of important IP, use Dropbox, iCloud, and local disks.
Share IP using Slack, Discord, iMessage, Apple Mail, or Signal.
The best AI app is Claude; ChatGPT is ok; don’t use Gemini, Meta, DeepSeek, or Grok.
GENERAL PRINCIPLES
Generative AI shouldn’t be used for final creative work, e.g. composing lines in a screenplay.
Writers will be paid to write, will own the copyright, and will be credited.
Semi-creative tasks might benefit from AI with full disclosure.
For semi-creative work, all parties need to opt-in to using AI at the start, with the default being no-use unless discussed.
Rule of thumb: if an AI would need to be credited, it has been used for too creative a purpose.
Creators’ IP should never be given to Big Tech companies for free.
AI models should not be trained on our creatives’ work without compensation.
Administrative or office tasks should be done with AI if it doesn’t hinder quality of work.
Feedback on this policy is welcome! Especially from writers.