AI Disclosure
Last updated May 31, 2026.
Where AI appears
Pocket Stoa uses a large language model — Anthropic’s Claude — in exactly one place: the Socratic Dialogue mode. Every other piece of content in the app is human-authored: the quiz questions, the dilemmas, the trolley cards, the maxims, the thought experiments, the chapters of The Last Day, and the philosopher commentary. None of it was generated by an LLM.
What the LLM does in Socratic
In Socratic Dialogue, the LLM plays one of five fictional characters who each defend a specific belief. The character’s personality, biography, belief, and the contradiction that would make them concede are all human-authored — they live on our server, not in the model. The LLM is, in effect, an improv actor reading from a tight brief. A second, very small LLM call checks whether the character has actually conceded the central belief.
What we don’t do
We don’t train any model on your conversations. We don’t send your text to third-party analytics. The Anthropic API does not retain your messages for training per their stated policy. Conversations are stored in your account only — visible to no one but you.
The floor
LLMs make mistakes. They will sometimes misattribute a quote, fabricate a citation, or give a confident-sounding answer that’s wrong. Treat the Socratic characters as imperfect interlocutors — sharp enough to be useful, not authoritative.