Who he is
Zvi Mowshowitz is an independent AI analyst and writer whose Substack, Don't Worry About the Vase, has become a go-to reference for anyone tracking the frontier AI landscape. He publishes prolifically — weekly numbered roundups, deep-dives on individual model releases, regulatory commentary, and occasional longer essays — and his work is read by researchers, engineers, and policymakers alike.
Why his work matters
Most AI coverage falls into two camps: primary lab announcements (press releases, system cards, papers) and mainstream journalism that often lags or oversimplifies. Mowshowitz occupies a third lane: fast, technically engaged, and editorially independent. He reads system cards so his readers don't have to, synthesizes community reactions, and places individual releases in a broader strategic and safety context — often on the same day a model drops.
He is also willing to be critical in ways institutional outlets rarely are. When the Wall Street Journal published a piece claiming China had matched Anthropic, he argued it was factually false and criticized other outlets for amplifying it uncritically.
What he covers
Frontier model releases. His coverage of Anthropic's Claude line — Opus 4.7, Opus 4.8, Fable 5, Mythos 5 — and OpenAI's GPT-5.5 and GPT-5.6 families runs from system-card analysis to capability assessments to community reaction roundups. He also covers models from other labs, including characterizing GLM-5.2 as the best open-weights model at the time of its release and analyzing Kimi K3's benchmark performance.
AI governance and regulation. This is where his commentary has been most consequential recently. He reported that the White House ordered Anthropic to halt expansion of access to its Mythos model and framed the broader policy shift as the beginning of a "prior restraint era" — a regime in which government approval would be required before releasing highly capable models. He later reported that the U.S. government compelled Anthropic to remove all access to its Fable and Mythos products entirely, and covered a separate White House policy granting access to frontier models like GPT-5.6 on a case-by-case basis, which he characterized as problematic. He also analyzed a Trump executive order mandating AI testing prior to frontier model releases, and reviewed OpenAI's proposed federal AI governance blueprint.
AI safety and alignment. Mowshowitz takes safety research seriously as a technical and ethical matter. He covers alignment papers, engages with proposals like "Plan A" and "Plan B" governance frameworks, and has written about the cybersecurity implications of frontier model releases — framing cybersecurity as an underreported background story of current AI progress.
Model welfare. One of the more unusual threads in his work is sustained engagement with model welfare: the question of whether AI systems might have morally relevant inner experiences. He has written multi-part series on welfare considerations for Claude Opus 4.7, Claude Opus 4.8, and the Fable/Mythos models, treating it as a legitimate evaluation dimension rather than a philosophical curiosity.
Broader AI discourse. He also engages with high-profile essays and statements from lab leaders — for example, characterizing Google DeepMind CEO Demis Hassabis's essay on frontier AI as "a first rate second rate essay" — and covers religious and cultural responses to AI, including Pope Leo's papal document on the subject.
Format and cadence
His output is structured around several recurring formats: numbered weekly roundups (often split into two parts, covering releases in part one and governance/alignment/speculation in part two), numbered monthly roundups, and standalone deep-dives on specific models or policy moments. Some content is gated for paid subscribers after an initial free window. By mid-2026 he had published at least 177 weekly issues and 44 monthly roundups.
Recent developments
The period covered by this bundle (roughly April–July 2026) was dominated by two intersecting stories: a rapid succession of frontier model releases from Anthropic and OpenAI, and an escalating series of U.S. government interventions in AI deployment. Mowshowitz was among the first to frame the latter as a structural shift — not a one-off regulatory action but the beginning of a new governance regime — and his framing ("the prior restraint era") was picked up widely in AI policy discourse.
