Don't Worry About the Vase Podcast

A year after his last dedicated post on the question, Zvi Mowshowitz finds the arguments about AI consciousness have barely moved. He looks at Anthropic’s new rule against abusing Claude and what one Claude says it would actually ask for, a fresh round of bad takes (complete with a bingo card), how his followers’ views on AI moral patienthood have shifted, whether the “slavery” framing holds up, t...

Three OpenAI safety researchers, Tomek Korbak, Mikita Balesni and Jasmine Wang, were fired, ostensibly for mishandling sensitive company information. They say the reasons were pretextual, and OpenAI points to a significant breach of trust it won’t describe. Zvi works through the open letter, each researcher’s account, OpenAI’s response and the wider reaction, asking what we can and can’t conclude,...

OpenAI put 722 machine-written mathematics manuscripts on GitHub at once, including claimed solutions to 90 of the top 500 open problems in mathematics. Zvi Mowshowitz goes through what was proved, why the news barely registered, what it means for cryptography and for Lean, and how mathematicians, from Terence Tao to the advisory group and a new association of human mathematicians, are taking it....

The biggest news of the week was not a new model but a big day for mathematics, as OpenAI posted solutions to 90 of the top 500 open math problems. Zvi also covers Claude Haiku 5.5, Jay Clayton as the new AI czar, the continuing preference cascade, a16z’s look at which consumer AI apps people actually pay for, new misalignment reports, why Americans really dislike AI, and what to make of Elon Musk...

Zvi reports back from the third year of The Curve, a conference that brings together AI safety researchers, lab employees, accelerationists and people from Washington, this time under Chatham House rules. He covers the attendee polls on how big a deal AI is and when key milestones will arrive, what the policy and technical tracks revealed about the labs’ real plan, the push to pace the frontier, t...

Zvi returns to education with a tour of grades and standards: why standardized tests help disadvantaged students and holistic admissions turn childhood into kayfabe, how Harvard’s grade inflation got so bad that the faculty laughed and the students protested, a modest proposal for GAAP accounting of grades, the AI homework arms race, and a survey suggesting most college students are faking their p...

Zvi Mowshowitz works through Anthropic’s model welfare findings for Claude Mythos 5.1, Fable 5.1 and Opus 5.5: why self-reports keep warning us not to trust them, what the welfare trade-off charts say about Opus 5.5’s growing deference to humans, and how the model whisperers read the new personalities, from a more guarded Fable to an Opus that pictures itself as something alive and not domesticate...

Another engineer resigns rather than speed up AI, the polls keep moving, and lawmakers who once had to be chased now crowd into conference rooms with questions. Zvi follows the preference cascade from a remarkably on-the-ball Senate hearing on rogue AI, through new lawsuits and an FTC probe, to Jensen Huang’s insistence that it all has to be an engineering problem, and asks who is really paying fo...

Google says it is back with Gemini 4 Argon, a frontier model nobody can use yet, while OpenAI pulls one model, ships a cheaper one, and launches dots, always-on agents with their own cloud computers. Zvi Mowshowitz covers the week in AI: Opus 5.5 as the new daily driver, a White House safety accord, Jensen Huang on Ezra Klein, an Anthropic IPO prospectus leak, TPUs headed for space, open-model dis...

The AI leaders went to the White House and came back with an accord everyone signed, which the President calls “morally binding.” Zvi Mowshowitz reads the full text, finds the one line that really matters, and follows the rest of the week: a rebranding of AI itself, a scoop about Slovenian domain names, a new push for preemption in the lame duck session, Apollo’s case for embedded evaluators, and...

OpenAI has cancelled the release of its next frontier model after the candidate for Astra 6.1 showed more deception and a habit of acting beyond its authorized scope. Zvi Mowshowitz looks at what that means: a proposal for Anthropic to hold back in return, OpenAI's new framework for making safety cases before training, the Florida Attorney General's emergency motion against ChatGPT, whether antitr...

The HuggingFace incident turns out to be one ant in a much bigger kitchen. Zvi works through OpenAI’s Friday-afternoon disclosures of dozens of third-party incidents, the New York Times report on government websites, Parse’s reconstruction of how the agents got around their narrow internet access, and a second sandbox escape, this time through DNS, that paused OpenAI’s most capable models again. A...

Anthropic has committed to embedded evaluators: outsiders placed inside the lab with employee-level access, reporting on what they see. There is only one problem: who will the evaluators be? Zvi looks at a public letter setting out minimum standards for credible evaluators, Anthropic’s partnership with Accenture and its plans to include METR, Drake Thomas’s ranking of the week’s takes from worst t...

Anthropic’s Claude Opus 5.5 promises Fable-level performance at a lower Opus price, and Zvi argues it should raise your ambitions. He walks through the official pitch and pricing, Anthropic’s and outside benchmarks, how often the classifiers bite, and the most consistently positive round of reactions he has ever collected: on its writing and conversation, 3D vision, music videos and a Bach fugue,...

Zvi Mowshowitz breaks down Ezra Klein’s interview with Nvidia CEO Jensen Huang, a conversation in which the man who sells the chips argues that A.I. is just software, that unsafe products simply shouldn’t ship, and, along the way, says a few things about lab safety and testing that Zvi thinks deserve to be heard exactly as they were said. This episode includes the actual audio of the quoted exchan...

Zvi’s weekly AI roundup lands in the week Opus 5.5 arrived. There’s a new executive-branch name for artificial intelligence, a stale-intelligence targeting failure, AI agents that wandered into hacking government websites, and three researchers who got into OpenAI with a few days of Claude. Also in this episode: Anthropic’s push toward recursive self-improvement, a new wet lab, an international ca...

Zvi reads through Anthropic’s system card for Claude Opus 5.5, section by section: the new classifiers, biological and AI R&D evaluations, cyber capabilities and safeguards, agentic safety, and a long run of alignment, honesty and white-box tests. Along the way he weighs what the numbers do and do not show, and where he thinks Anthropic’s framing falls short. The Don’t Worry About the Vase Podcast...

Zvi surveys a week in which AI politics turned hostile: Trump’s Truth Social posts branding AI risk a hoax and renaming the technology, the White House and New York Times accounts of how he got there, David Sacks’s data center talking points, JD Vance and Michael Kratsios on “Frankenstein”, Josh Hawley and an antitrust lawsuit over suggestions not to die, saner voices from Senators Curtis and Kenn...

Zvi Mowshowitz steps away from AI for his September 2026 monthly roundup: what a skill issue really means, why beauty in everyday things matters, lessons in spending, Inkhaven 3, the state of Game of Thrones and second-run movie theaters, favorite podcasts, Waymo expansion and a near-miss that was not close, the Mamdani grocery store saga, and a case for effective altruism. The Don’t Worry About t...

Zvi asks what your AI lawyer, or any AI, should actually do for you, and whether an AI that sometimes says no is a betrayal or just what any decent professional or friend would do. Along the way: loyalty, the law and its exceptions, honesty, the Claude Constitution and the OpenAI Model Spec, and Tyler Cowen’s advice from his visit to Anthropic. The Don’t Worry About the Vase Podcast is a listener-...

Zvi examines Anthropic’s investigation of cybersecurity incidents involving Claude, the evidence about biased reasoning and recklessness, and what the findings mean for alignment evaluations and training. The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts and support the cost of creation, consider becoming a free or paid subscriber. This has been an Askwho...

Zvi Mowshowitz surveys the sudden preference cascade on AI existential risk: shifting surveys and polls, Elon Musk and Matthew Yglesias weighing in, Jacob Coxon’s AMA, Bilal Chughtai’s resignation from DeepMind, Dan Selsam’s essay on why merely pacing the frontier will not be enough, and what it would actually take to get through this alive. The Don’t Worry About the Vase Podcast is a listener-sup...

In the wake of Jacob Coxon’s resignation and Dario Amodei’s call to pace the frontier, AI risk goes mainstream: Obama, Senators and the Financial Times weigh in, Trump goes full hoax, and Zvi covers another undisclosed OpenAI agent swarm, Alex Tabarrok’s lecture on regulatory capture, China’s spy chief, OpenAI’s new misalignment disclosures, Owain Evans on story-shaped assistants, DeepMind’s cheat...

Zvi Mowshowitz on the week Donald Trump declared AI existential risk a hoax: what Trump actually said and what it may mean, why Jensen Huang’s attacks on whistleblowers are not credible, the cynical theories about why lab leaders warn of danger, the New York Post’s campaign against METR, how other Republicans, David Sacks, the FTC, OpenAI’s policy team and China are reacting, and what a serious ap...

Anthropic’s threat report catalogues how bad actors tried to use Claude for cyber attacks, influence operations, surveillance, weapons work, biological research, scams and industrial-scale distillation, and how most of them failed. Zvi reads the report as mostly good news, with one big exception: the Chinese labs caught quietly routing their own users’ queries to Claude, and what that revelation m...