Three Labs With a Plan and A Memorandum

TL;DR

• 1

  OpenAI explicitly called for the ability to slow frontier AI: Sam Altman and Jakub Pachocki said there should ultimately be an international organization that can coordinate leading labs and even slow development when catastrophic risk, safety, and alignment demand it.

• 2

  The White House memorandum looks like an Anthropic ban in all but name: NSPM-11 says agencies should terminate contracts, including subcontractors, with companies whose behavior conflicts with the memo, with only limited one-year waivers that the host reads as a direct response to the DoD-Anthropic fight.

• 3

  The key fight is over control, not just competition: The host argues OpenAI focuses on concentration of power while Anthropic focuses more on diffusion and loss of human control, and says those two risks cannot be cleanly separated once recursive self-improvement is on the table.

• 4

  OpenAI's plan openly centers recursive self-improvement: The company says AI doing AI research could set the pace of progress within a few years and that by March 2028 a significant fraction of its own research may be done by AI systems working alongside researchers.

• 5

  Anthropic staff pushed back on Joshua Achiam's 'machine god vs human tools' framing: In a poll and quoted replies, employees like Amanda Askell and Sarah Chen said the binary is false and that many at Anthropic see 'Culture-style' AI stewardship as a disempowering outcome, not the goal.

• 6

  The policy language on 'assurance' is where the memo gets sharp: The memorandum says no commercial entity can prevent, disable, degrade, or materially modify AI systems the government relies on, which the host translates as: if the state wants to ignore your terms of service, it can.