Zvi Mowshowitz, who writes the analysis blog Don’t Worry About the Vase, published his own reading of the frontier lab pacing letter on Wednesday, and his count of signatures has climbed to 1,224 employees. That is higher than the figure Axios reported when AI Insiders covered the letter’s launch, and it now carries formal endorsements from both OpenAI and Anthropic as organizations, not just individual staff.
The letter itself asks the U.S. government to back an international effort to build technical and governance tools for slowing frontier AI development on purpose, if that ever becomes necessary. It does not call for anyone to slow down today. Mowshowitz is careful to keep that distinction intact: the request is for the option to pace progress later, built now while there is still time to build it.
Mowshowitz’s own conclusion goes further than summary. He argues the letter earns its signature count because of two specific choices its drafters made: separating the act of preparing coordination tools from any demand to use them immediately, and swapping charged words like pause or shutdown for the milder “pace.” In his reading, that wording let executives who disagree on how urgent the risk is still sign the same document, which is why Dario Amodei, Jakub Pachocki, and dozens of other lab leaders appear on it.
He also passes along a signature breakdown compiled by researcher Andrew Trask that shows how unevenly the letter landed inside each company. By that count, roughly 9.8 percent of Anthropic’s staff signed, versus 3.3 percent at OpenAI and 1.9 percent at Google DeepMind. Mowshowitz notes that xAI is conspicuously absent from the list, without offering a reason why.
Mowshowitz is not uncritical of the document. He singles out its opening line, that AI “could help create a dramatically better future, but that outcome is not guaranteed,” as the most watered-down sentence in the whole statement, and says he was disappointed it was not written with more force. He largely aligns with a harsher, clause-by-clause takedown from MIRI’s Nate Soares, agreeing that phrases like “develop security measures” and “strengthen oversight” let readers assume normal safeguards will be enough without the letter ever claiming that outright. Mowshowitz pushes back on Soares in a few spots, arguing some of his objections read the letter too literally, but he ends up close to Soares’s bottom line: a genuine step forward that still falls short of the direct language the situation calls for.
The distinction matters for how operators should read this news. The letter’s substance did not change between its launch and Mowshowitz’s piece; what changed is the count, the two formal lab endorsements, and now a public argument among safety researchers over whether its hedged language was a necessary concession or a missed opportunity. Mowshowitz’s own position, that the wording tradeoff was worth it to get OpenAI and Anthropic both on record, is itself a bet that has not been tested yet.
That bet gets tested the next time a lab ships a model capable enough to force the question the letter only prepares for. Anyone tracking frontier safety commitments should watch whether the coordination infrastructure the letter requests actually gets built before that moment arrives, not after.
Zvi Mowshowitz analyzed the pacing letter on his blog, Don’t Worry About the Vase, on July 29, 2026.
