Emad Mostaque

Profile reviewed Oct 14, 2026

• Intelligent Internet introduction
• Stability AI departure announcement

Source published Jul 24, 2025

• Master Plan (open source, not so secret 🤷🏾)

Source published May 16, 2023

• Advocating for Open Models in AI Oversight: Stability AI's Letter to the United States Senate

Source published Aug 14, 2024

• The Sequence Chat: Emad Mostaque -Stability AI, Schelling AI- About Open and Decentralized AI

The page displays 24 Jul 2025 above the byline naming Emad Mostaque, Founder and CEO. Opening diagnosis and the four-part plan, especially sovereign AI

Full short essay retrieved from the publisher page.

The plan proposes a benefit-backed ledger, user-owned agents, a coordination protocol, and auditable shared knowledge. Those mechanisms are proposals.

A manifesto. It does not show that the proposed ledger or agents exist or distribute power. Company advocacy that promotes Intelligent Internet's own products (II-Agent, II-Medical, Common Ground).

The continuity is the objection to centralized opaque control. The new piece adds an ownership plan. It is not a retraction of his scaling answer.

Editorial record as of Oct 14, 2026.

Interview header reads Aug 14, 2024. Rapid-fire answer to How far can the LLM scaling laws take us?

Original question-and-answer text read. The biography and questions are the publisher's. One earlier answer is cut off on the page and was not repaired.

The unlikely route is AGI from scaling language models. He separately says broader systems of humans and generative models are harder to predict.

A short interview answer, not a technical paper. The publisher's biography is not his claim.

This is a claim about language-model scaling, not about who should control deployed systems. That access argument is a separate answer in the same interview.

Editorial record as of Oct 14, 2026.

Interview header reads Aug 14, 2024. Answers on decentralized training, open versus closed models, and public-sector black boxes

Same retrieved interview. The broken sentence in the first answer was not used.

He separates centralized training of base models from later customization and says the goal is useful distribution, not decentralization for its own sake. He calls for new organizations to fund public AI infrastructure and expects open models to be widely used if they remain good enough.

Interview answers, not a statute or a system design. The always-outperform remark rests on his reasoning that closed developers can add private data to open models, not on a benchmark.

Same day as his answer on language-model scaling. The performance and training remarks have no counterpart in his 2023 letter, so they are new topics, not a revision of it.

Editorial record as of Oct 14, 2026.

The announcement page's datePublished metadata is 2023-05-16T16:48:22+0000. The signed letter itself is dated May 13, 2023. The public release date is used. Signed cover letter, PDF pages 1–2

The signed letter and attached company submission were read from the publisher PDF. The announcement page's dateModified is 2024-10-14, so the current page is not assumed to be an untouched May 2023 snapshot.

The letter ties openness to public scrutiny, competition and US leadership, and says public oversight starts with public scrutiny. It cites his signature on the March 2023 open letter urging caution and Stability AI's participation in community-led model testing announced by the White House.

Written mostly in Stability AI's voice and signed by Mostaque as CEO. The attached risk taxonomy and five oversight suggestions are the company's submission. The announcement page names a different CMS author and was modified on 14 October 2024; the signed letter itself is dated 13 May 2023.

This is his signed request for openness. It does not say that every model must be trained in a decentralized way or that open models will match closed ones.

Editorial record as of Oct 14, 2026.