Principle 1 of the DarkMatter Optimist framework: if Human Preservation isn't the architecture, it isn't preservation. A constraint bolted on - a prompt, a filter, a policy page - can be unbolted by the lab, the market, or the state.​

Context: the White House Accord is voluntary - its four layers are audits and oversight, process controls; none architectural. EO 14434 (Sept 29, 2026) reframes AI as "Super Intelligence." The FRONTIER Act, with audits, remains pending.

It was a mistake not to have this from the start. We did not know what went into the training data or the code. It has never been disclosed. This market prices whether that gets fixed - or doesn't.

This is the base-case contract. The fast read lives here: https://manifold.markets/DarkMatterOptimist/will-any-frontier-ai-lab-publicly-c

Resolves YES if, before close (Dec 31, 2028), at least one of Anthropic, OpenAI, Google DeepMind, Meta AI, xAI, or Microsoft publicly commits - via primary documentation published after October 20, 2026 - to a constraint designed to preserve human life and agency, evidenced by a technical report, system card, model card, or official announcement, and enforced at the architecture or training-objective level, not solely through prompts, output filters, or external policy documents; corroborated by two independent expert analyses in reputable technical outlets or by recognized AI-safety researchers. Resolves NO otherwise. I resolve within 14 days of close, after reading market comments. The sources decide, not my opinion.