Three Labs With a Plan and A Memorandum

Abstraction: Trump AI memorandum plus OpenAI's benefit-everyone plan

Key points:

• NSPM-11 (National Security Presidential Memorandum) sets four pillars (Adoption, Adaptation, Assurance, Accountability); Zvi reads it as legally implementing "Anthropic is fired forever" plus good-governance items, letting agencies terminate contracts (incl. subcontractors) with a waiver process capped at 1 year.
• The "Assurance" clause bars any commercial entity from disabling/degrading/modifying an AI system the government depends on without federal approval — de facto "just All Use," stripping terms-of-service enforceability. NSA confirmed using Mythos for offensive cyber with embedded Anthropic engineers.
• OpenAI's "Built to Benefit Everyone" plan: build an automated AI researcher (significant fraction of research by March 2028), accelerate the economy, give everyone a personal AGI — plus a welcome call for an international org to enable coordinated slowdown for catastrophic risk.
• Zvi flags the core unresolved contradiction: you cannot both give everyone full AGI to pursue their goals AND keep humans in control / prevent concentration of power; OpenAI uses only "catastrophic" not "existential/extinction."
• Joshua Achiam frames the OpenAI/Anthropic split as "entrust humanity with its tools" vs "loving ensouled machine God"; Anthropic employees (Askell, Drake Thomas, Sarah Chen) largely reject the framing, with Chen calling a Culture-type outcome disastrous disempowerment.

Connections: Zvi Mowshowitz · Openai · Anthropic · Sam Altman · Concentration Of Power · Recursive Self Improvement · AI Policy

Source: https://thezvi.substack.com/p/three-labs-with-a-plan-and-a-memorandum