Michael Darius Eastwood · Independent AI alignment researcher
Published
Embedded Off-Switches for AI Compute: thousands of deadman-switch security blocks per accelerator, <1% die overhead.
James Petrie's embedded off-switches paper proposes thousands of independent security blocks per AI accelerator, each a deadman's switch requiring cryptographic authorisation to keep the chip alive, at under one percent die overhead. Published September 2025, roughly four and a half months after the manuscript that described Meltdown Triggers: automatic fail-safes that deactivate a system attempting to bypass its ethical constraints. Same architecture, same threat model, independent derivation.
Source: arXiv:2509.07637.
Open the primary source above. Confirm the date. Confirm the finding. Then compare it against the manuscript passage or artifact named in the register entry, whose SHA-256 hash is published on the evidence page. Nothing in this article asks for trust; the entry either survives that comparison or it comes off the register, publicly, the way the programme's retracted scaling figure did.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.