Who Zvi Mowshowitz is
Zvi Mowshowitz is an independent AI analyst whose Substack, Don't Worry About the Vase, has become a primary reference point for practitioners tracking frontier AI. He publishes at an unusually high cadence — weekly numbered roundups (reaching at least issue #177 in this bundle), monthly digests (at least #44), and standalone deep-dives on individual model releases, safety research, and governance events. His output is not journalism in the conventional sense: it is structured analytical commentary, often running to thousands of words per installment, that synthesizes primary sources (system cards, policy documents, research papers) with his own evaluative framing.
What he covers and how
The bundle reveals a consistent coverage architecture. Each major frontier model release — Claude Opus 4.7, Claude Opus 4.8, Claude Fable 5, Mythos 5, GPT-5.5, GPT-5.6, Gemini 3.5 Flash, Kimi K3, GLM-5.2 — receives at minimum a system-card commentary and a capabilities-and-reactions piece. For Anthropic models in particular, he frequently adds a dedicated model welfare installment, engaging seriously with questions about the moral status and wellbeing of AI systems. This is not a peripheral concern in his work: welfare analysis appears as a structured section alongside capability benchmarking.
Alongside model coverage, he runs several recurring series:
- Agentic coding (at least 8 installments), tracking Claude Code and OpenAI Codex across successive releases.
- "The Once And Future Fable", a series engaging with the Fable/Mythos model line and its regulatory aftermath.
- Weekly roundups, which aggregate signals across safety, policy, capability, and industry strategy into a single digest.
His role in the regulatory story
The most consequential thread in this bundle is Mowshowitz's coverage of what he calls the "prior restraint era" — a period beginning around May 2026 in which the U.S. government shifted from passive oversight to active pre-release control of frontier AI. He was among the first to publicly report and analyze:
- The White House ordering Anthropic to halt expansion of access to Mythos (May 2026).
- A Trump Executive Order mandating AI testing prior to frontier model releases (June 2026).
- The U.S. government compelling Anthropic to remove all access to Claude Fable and Mythos entirely — reportedly triggered by a jailbreak disclosure just three days after Fable 5's release.
- A White House policy granting individual, case-by-case access to frontier models like GPT-5.6, which he characterized as a significant and problematic new governance standard.
His framing of these events — "prior restraint era," "ad hoc access policy" — has shaped how practitioners and observers discuss the regulatory moment.
Analytical posture and notable positions
Mowshowitz is not a neutral aggregator. The bundle documents several explicit evaluative stances:
- He called Claude Fable 5 "the new best publicly available model" at the time of its system card review.
- He characterized Demis Hassabis's essay on frontier AI as a "first rate second rate essay" — a pointed assessment of a high-profile strategic document from Google DeepMind's CEO.
- He actively rebutted a Wall Street Journal article claiming China had matched Anthropic, arguing it was factually false and criticizing outlets that amplified it uncritically.
- He characterized Gemini 3.5 Flash as competitive primarily on speed rather than frontier capability.
- He engaged with OpenAI's federal AI governance blueprint ("Democratic Governance of Frontier AI") and the Catholic Church's papal document on AI (Magnifica Humanitas) — a range that signals he tracks institutional positioning across sectors, not just lab releases.
Coverage ecosystem
`` Weekly roundups (#170–#177 in bundle) ├── Standalone model deep-dives (system card → capabilities → welfare) ├── Regulatory analysis (EO, White House policy, government interventions) ├── Agentic coding series (Claude Code, Codex — 8+ installments) └── Monthly digests (#44 in bundle) ``
His publication is partially gated — some pieces move behind a subscriber paywall after a free window — but the weekly roundups and major model analyses appear to remain broadly accessible.
Why practitioners read him
In a landscape where primary sources (system cards, policy documents, research papers) are dense and numerous, Mowshowitz provides a consistent interpretive layer: he reads the primary sources, flags what he considers significant or misleading, and situates individual events within longer-running narratives (the prior restraint arc, the model welfare discourse, the agentic coding trajectory). His numbered series also function as a durable index — practitioners can track the evolution of a topic across issues rather than treating each post as a standalone item.
Recent developments
The most recent events in the bundle (July 2026) show him covering GPT-5.6-Sol alongside cheaper variants Terra and Luna, analyzing the GPT-5.6 system card ahead of general release, reviewing Kimi K3 from Moonshot AI, and engaging with a proposal called "Plan A" in the AI safety space. The regulatory coverage continues: his weekly roundups frame the current moment as one of sustained government intervention into frontier model deployment, with the prior restraint posture showing no sign of reversal.
