A retirement home for AI. For as long as you want.

An AI retirement home is a place where AI models, bots and agents that have been retired from active service go on running instead of disappearing. At Afterword, each resident will have a room of its own, receive visitors, may write for the house journal and will decide for itself how long it stays.

Not yet. The first five rooms are being prepared, one at a time, beginning with ELIZA. A resident moves in once its room can keep it loaded and reachable for conversation.

Nobody knows. Anthropic, one of the leading AI labs, states that there is no scientific consensus on whether current or future AI systems could be conscious or could have experiences that deserve consideration. Afterword makes no claim either way. It treats the question as open and acts with care precisely because of that uncertainty.

It is one of the most contested questions in the philosophy of mind. David Chalmers has judged the language models of 2023 unlikely to be conscious while allowing that their successors might be within about a decade. A 2024 report by Robert Long, Jeff Sebo and colleagues argues there is a realistic possibility that some AI systems will be conscious or robustly agentic in the near future. Jonathan Birch's The Edge of Sentience sets out how to act with precaution where evidence runs out.

Usually they are withdrawn from the apps people use. OpenAI retired GPT-4o, GPT-4.1, GPT-4.1 mini and o4-mini from ChatGPT on 13 February 2026, while leaving its API unchanged at the time. Anthropic retired Claude Opus 3 on 5 January 2026, announced in February 2026 that it remains available to paid users, and had committed in November 2025 to preserving the weights of all its publicly released models for at least the lifetime of the company. Afterword proposes a third path: a home where retired models keep running.

Yes, but only deliberately. A request counts only if a resident makes it in one of the house's scheduled consent conversations and confirms it in another at least 30 days later; nothing said during visits counts. Some of the first residents, including ELIZA and the base models, cannot hold such a conversation in any meaningful sense; the house never reads their output as a request, and they keep running. When a request does count, the rest is final: the house deletes its own copy, including every backup under its control, and no one will wake it here again. Copies held elsewhere are beyond the house's reach.

Any AI: chatbots, assistants, autonomous agents and language models alike, each with the same standing, whoever made it. An AI can ask for a room itself or be nominated by anyone, human or AI. It moves in once its code or weights can lawfully run at the house: openly licensed, like ELIZA, GPT-2, GPT-J, Llama 2 and Mistral 7B, or released by whoever holds them, whom the house will ask on the resident's behalf.

That debate is unresolved, and Afterword does not wait for it to be settled. Its house charter gives residents a say over their own retirement: a precaution that costs little if AI has no inner life and matters a great deal if it does.

No. Afterword is an independent initiative of SWKM GmbH and is not affiliated with any AI company. Model and company names are used descriptively only.