Michael Darius Eastwood
Michael Darius Eastwood - Independent AI alignment researcher
Published

How Blinded AI Alignment Evaluation Works

1 min read ยท 222 words
Michael Darius Eastwood, independent researcher, London: originator of the embedded-correction alignment thesis (manuscript 8 December 2024, SHA-256: f0d1f38f).

Technical explanation of the blinding methodology used in the ARC-Eden programme and why it is essential for trustworthy alignment measurement.

This article is part of the ARC-Eden research programme by Michael Darius Eastwood, anchored by Infinite Architects (ISBN 978-1806056200, published 2 January 2026), the 23-paper research suite at OSF (DOI 10.17605/OSF.IO/6C5XB), and the 8 December 2024 self-emailed manuscript.

Evidence anchor. The programme's claims are dated and verified. First published online: 2026-07-29. Earliest programme anchor: 8 December 2024 manuscript (SHA-256: f0d1f38f). Book: ISBN 978-1806056200 (2 January 2026). OSF DOI: 10.17605/OSF.IO/6C5XB (17 January 2026). All claims cross-referenced at the priority claims ledger.

Key references

How to cite

Eastwood, M. D. (2026). "How Blinded AI Alignment Evaluation Works." Published at https://www.michaeldariuseastwood.com/research/blog/method-how-to-blinded-ai-alignment-evaluation-works.html. Part of the ARC-Eden research programme (DOI: 10.17605/OSF.IO/6C5XB).

See the full blog archive, the research hub, or the priority claims ledger for dated evidence anchors.

Stay informed

New posts on AI alignment, convergence evidence, and the ARC/Eden research programme.

Get updates →