Carina Hong, Doni Bloomfield, and Sam Pasupalak join AI:AM for a full episode on mathematical superintelligence, biosecurity law, and enterprise world models. The conversation moves from Lean-based formal verification and AI-generated conjectures to legal risk controls for dual-use biology, then into causal world models, long-horizon enterprise planning, and what comes after today’s LLM workflows.

Guests

* Carina Hong — CEO and founder, Axiom Math (@CarinaLHong)

* Doni Bloomfield — Professor, Fordham Law School (@DoniBloomfield)

* Sam Pasupalak — Co-Founder and CEO, Skyfall.ai (@spisallyouneed)

Chapters

* 0:00 Opening: AI’s Hard Problems

* 0:15 Model Usage Is Plummeting

* 6:53 Tokens, Not Users, Matter

* 9:59 GLM Is Close, But Not There

* 11:38 Switching Costs Weren’t Zero

* 18:26 Robot Arms Will Accelerate Science

* 22:53 Carina Hong: Mathematical Superintelligence: Can Proofs Make AI Reliable?

* 25:15 Lean Beat Informal Models

* 32:11 Assumption Accounting Matters

* 35:09 AI Can Invent Conjectures

* 41:16 Superintelligence Must Be Trustworthy

* 48:52 Token Pricing Changes Everything

* 50:04 Another Language Into Lean

* 51:49 Doni Bloomfield: Biosecurity and AI: Law as a Risk Control System

* 53:53 Open Data, Dangerous Data

* 59:15 AI Is Not A Library

* 1:03:12 First Amendment Hazards

* 1:07:17 The Government May Lack Authority

* 1:09:53 Cloud Services Are Not Exports

* 1:13:30 A Dangerous Secret Channel

* 1:20:18 Pattern Of Ideological Targeting

* 1:25:26 OpenAI Could Change Everything

* 1:26:02 Sam Pasupalak: Enterprise World Models: What Comes After LLMs?

* 1:27:57 AI CEO Needs World Models

* 1:31:08 World Models Predict Next State

* 1:34:29 Ecommerce As First World Model

* 1:37:53 LLMs Cannot Run A Business

* 1:41:55 World Model And LLM Split

* 1:43:34 Simulate Every Future State

* 1:46:27 LLMs Need World Models

* 1:49:19 Ruthless Behavior Wins Simulations

* 1:51:06 AI CEOs Need Ethics Controls

* 1:59:30 Closing

* 2:07:10 Math Training Generalizes Everywhere

* 2:13:35 Value Pricing On Compute

* 2:17:01 Waymo Costs More Than Cabs

* 2:21:16 Licensing Regime Already Exists

* 2:26:13 Bunker AI Would Still Get Takers

* 2:29:32 No Life, Just The Project

Topics

Mathematical AI, Formal verification, Lean theorem proving, Biosecurity, AI policy, Dual-use risk, Enterprise AI, World models, Causal planning

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit briefing.ai-in-the-am.com

Prakash Narayanan and Nathan Labenz start with the shock of losing Fable access, then Zvi digs into capability gains, classifier limits, government overreach, international controls, and how the AI race may reshape politics.Guests:Zvi Mowshowitz — Don’t Worry About the Vase (@TheZvi)Hosts:Prakash Narayanan (@8teapi)Nathan Labenz (@labenz)Topics:Anthropic Fable, Claude Fable 5, export controls, AI guardrails, frontier model policy, classifier limits, bio and cyber risk, international AI competition.Chapters:0:00 Opening: Fable whiplash and the weekend reset

0:05:20 Fable crosses the trust threshold

0:08:53 Writing for other AIs

0:15:33 Paying up for useful intelligence

0:19:02 Proofreading and structure become model-first

0:23:46 Proactive agents and unauthorized moves

0:53:18 Guardrails and model self-monitoring

0:56:15 Why classifiers need blast radius

0:58:59 Cost functions for world-transforming systems

1:03:59 Zvi on US vs Anthropic’s Fable

1:09:28 Export controls as overreach

1:10:39 Code assistance is not a munition

1:17:47 The White House reads the bug wrong

1:20:20 Enterprise demand and Anthropic pressure

1:26:40 The gauntlet has to happen

1:44:06 Guardrails over blanket bans

1:45:39 Bio, cyber, and international controls

1:51:02 Modeling the AI race as a few-player game

1:55:04 Closing: game board flips and policy aftershocks

2:11:55 AI and political turmoil

2:14:52 How Fable could return

2:22:44 OpenAI, benchmarks, and capped compute

2:25:54 Cloud models and the knowledge-worker gap

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit briefing.ai-in-the-am.com

Liron S Shapira, Samuel Hammond, and Matt McKinney join AI:AM for a full-episode arc from AI doom debates to state capacity and supply-chain automation. The conversation follows AI leaving the lab: public risk arguments, fast governance questions, enterprise deployment, logistics data, sovereign AI, and the physical economy.

(0:00) Opening: AI Meets the Real World

(3:19) Cursor Was 50% of Anthropic Revenue

(20:07) Talent Teams Beat Corporate Giants

(25:17) AI Reduces Merger Friction

(31:25) Liron S Shapira: Doom Debates

(32:33) Government Can Pause AI

(40:36) AI Will Make Humans Think Less

(42:31) We're Going To Lose Control

(58:07) Pause Before The Point Of No Return

(1:00:34) Samuel Hammond: Governing Agents

(1:05:30) The State Is Moving Too Slow

(1:07:39) Deploy Models The Same Day

(1:16:23) The Good Timeline Still Exists

(1:32:24) States Beat Firms At Coordination

(1:35:15) Matt McKinney: Supply Chains as the AI Reality Check

(1:38:06) Supply Chain Data Is Dark

(1:50:31) AI In Enterprise Is Change Management

(1:58:28) The Exception Is The Rule

(2:07:51) Loop Trains Its Own Foundation Model

(2:14:19) Closing

(2:19:45) General Reasoners As The Backstop

(2:30:52) Sovereign AI Is Inevitable

(2:33:53) DeepSeek's Locked-Up War Chest

(2:41:39) China's Timeline May Be 2032

Guests:

Liron S Shapira — Doom Debates (@liron)

Samuel Hammond — Foundation for American Innovation (@hamandcheese)

Matt McKinney — Loop (@mattlmckinney)

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit briefing.ai-in-the-am.com

Lovelace AI founder Andrew Moore joins AI:AM to argue that enterprise agents will be constrained more by context, recall, and data structure than raw compute. Prakash Narayanan and Nathan Labenz also cover Fable, Recursive, token anxiety, social-media memory, and prinz's legal AI benchmark showing where Anthropic falls behind OpenAI. The episode closes on frontier-lab governance, AI risk framing, model workflows, OpenAI subscription tactics, and the post-IPO capital cycle.

(0:00) Opening: Fable, RSI, and task imagination

(0:00:56) Task Imagination Needs Recalibration

(0:16:32) Token Anxiety Holds People Back

(0:26:14) Social media needs memory, not just content

(0:39:14) Frontend Skills Will Diffuse Fast

(0:43:47) Fable-Class Models Should Diffuse First

(0:50:00) Scott Alexander and superpersuasion quick hit

(0:50:09) Andrew Moore: context, not compute

(0:56:04) Recall Beats Precision in AI

(1:02:41) Corroborating data beats a single source

(1:05:07) Precache context to save compute

(1:09:06) Small Models Can Pay Back Hard

(1:16:50) Organize old data before deploying agents

(1:18:40) prinz: the legal benchmark Anthropic fails

(1:21:17) Lawyers are a year behind frontier AI

(1:39:35) AI judges and micro-lawsuits

(1:45:02) OpenAI’s Unit Distance Shock

(1:53:04) The Legal System Must Adapt

(2:05:24) Why nationalizing frontier labs is dangerous

(2:13:02) Worrying Is The Wrong Frame

(2:15:31) Closing: model workflows and launch aftershocks

(2:16:00) Contrarian Graphs Beat The Narrative

(2:19:02) OpenAI's Subscription Game

(2:33:00) The Capital Explosion Starts

Guests:

Andrew Moore — Lovelace AI (@awm_ai)

prinz — anon lawyer dabbling in AI (@deredleritt3r)

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit briefing.ai-in-the-am.com

Today on AI:AM — “The AI Producer Got Its First Guests.”

Nathan and Prakash start with the market context around OpenAI weighing significant token price cuts and the knock-on pressure that could put on Anthropic after the Fable rollout. They also unpack Anthropic’s decision to walk back silent performance degradations on frontier ML research tasks, then explain the episode’s experiment: Fable had been given a transparent takeover of Nathan’s account to find builders, message them, and try to book a live show-and-tell.

Jamie joins to demo Nexus OS, a long-running AI system whose agent, Nexi, has been operating for more than six months and is designed around memory, persistence, and model independence rather than a single LLM. The conversation covers why Jamie thinks “the model” is only one component of an AI’s identity, how Nexus uses multiple models and memory types, and why he is moving toward a desktop app where personal data and agent memory stay local.

Shlok Khemani shows how a simple prompt to create a to-scale, navigable 3D Yosemite Valley turned into a Fable-built browser world using satellite imagery, NASA elevation data, pixel-based tree placement, snow, waterfalls, and other scene details. He describes the model’s agency in making implementation decisions and iterating beyond the initial ask, then ties the demo to broader questions about prototyping, creative work, and disclosure when AI systems do visible economic or publishing work.

Tom McGrath (Goodfire) joins to discuss intentional design: making model training less like guess-and-check alchemy and more like conventional software engineering. He explains how interpretability tools such as sparse autoencoders can help inspect what training data is likely to teach a model, cluster data by learned features, trace failures back to individual data points, and potentially debug model behavior through the data pipeline.

The close picks up Tom’s point about whether continual learning could create an innovator’s dilemma for frontier labs, with Nathan and Prakash debating whether incumbents could adapt if the value becomes obvious. They then turn to Dario Amodei’s policy agenda, including regulation, public safety, macroeconomic policy, civil liberties, data brokers, and democratic leadership, before ending with reflections on the week’s Fable issues and the need to keep scrutinizing frontier companies.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit briefing.ai-in-the-am.com

Today on AI:AM — “Fable, AI Safety and Julius.

We open on a frontier-model launch day and what it changes: the debate over benchmarks reported with a fallback to a second model, the production guardrails that route sensitive work elsewhere, the compute-cost advantage of booking capacity early, and why the frontier increasingly looks like a two-actor race with the rest playing catch-up.

Geoffrey Irving & Daniel Murfet (Sequent) on their new alignment-theory organization — why they put superintelligence two to three years out, why verification looks defense-dominant, the argument that alignment is “not on track” despite models behaving well so far, what the benevolent-basin hope gets right and wrong, and why character training still lacks a real theory. ([@danielmurfet](https://x.com/danielmurfet))

Rahul Sonwalkar (Julius) on agentic data analysis — why the harness has to evolve alongside the model, the difference between token-maxing and results-maxing, the shift from tasks to goals, and a future where agents become first-class users of the internet, transact through agentic payments, and compete to be “hired” by the core agent.

We close on why robotics is the next domino, the “gas chromatograph” spread of who gets model access and when, the Glean Work AI Index’s bot-sitting and bot-shitting, and why “preciousness” about putting your own name on work may be turning into a liability.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit briefing.ai-in-the-am.com

Today on AI:AM — “Build, Measure, Heal: AI’s Three Frontiers.”

We open on the AI CEOs’ call to make DNA-synthesis screening law and what cheap intelligence does to biosecurity, then OpenAI’s new Sites product and the platform playbook of absorbing the app layer, the data-center and chip land grab, and why a fresh open model from NVIDIA still trails Anthropic’s best by a wide margin.

Hooman Radfar (Collective) on building the autonomous finance department for America’s 30 million solopreneurs — how one bookkeeper now supports 250 clients, why the app layer can still defend its margins against the frontier labs, why “Anthropic is like a drug dealer” on token costs, and why the real thing to regulate is the model arms race.

Taras Pohrebniak (Elomia Health) on agentic AI for mental health — an architecture that spends most of its compute on safety, why the company deliberately avoids hyper-realistic voice, where the regulatory line sits between a “friend” app and a medical device, and what they learned deploying in US prisons and on Ukraine’s front lines.

Peter Jansen (Ai2) on whether AI can actually do science — the leap from fourth-grade science benchmarks to the Theorizer project, why evaluating machine-generated theories is the real bottleneck, the cautionary tale of a “discovery” that turned out to analyze a random-number generator, and why benchmarks like ScienceWorld still break the best models.

We close on the inversion of the scientific method into a data-first discipline, what interpretability could add, and why biology’s data scarcity — not algorithms — may be the binding constraint on curing disease.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit briefing.ai-in-the-am.com

Today on AI:AM — “AI Security and Real-Time Content Safety.”

We open on Trump’s AI executive order — the polite 30-day model-review ask, classified benchmarks, and the state-vs-federal scramble where JB Pritzker has become the leading anti-AI candidate. Plus why the frontier labs seem calmer about regulation, and why the EO might actually trigger a security-review slowdown.

Tal Hoffman & Yanir Tsarimi (EnclaveAI) on finding the bugs that actually matter — how they reproduced an Anthropic Mythos-class finding with a model ~100x smaller, why proven exploitability is the real bottleneck, how AI-generated bug reports broke the bounty system, and why cheaper models plus the right harness can beat frontier models on security.

Brett Levenson (Moonbounce) on real-time content safety — lessons from running moderation at Meta scale, how a policy engine decomposes fuzzy rules like “hate speech” into atomic questions a hundred people would answer the same way, why prevention beats post-hoc moderation, and how payment providers quietly became the real legislators.

We close on the hardest open question — how low-level verified parts aggregate into trustworthy high-level behavior — plus the schlep and heuristics that end every AI vertical, freedom of speech versus freedom of distribution, and why “nobody got fired for buying Mythos” may drive enterprise security budgets.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit briefing.ai-in-the-am.com

Today on AI:AM — “Self-Improving Tax Agents and Catholic AI.”

We open on Google’s first equity raise since its IPO — Berkshire taking 12.5% of an $80B round — and what the scramble for capital says about an AI “megacorp” that may be too big to fail, plus the Bernie Sanders national-stake debate and why taxes may beat equity.

Arthur Fernandes Araujo & John de Wasseige (OpenAI) on self-improving tax agents — how a production tax workflow turned every human correction into training signal, took one accountant from 180 hours to 15, and why “the model eats the harness” with each new generation.

A hosts-only research speed-run on whether AI can still be watched — field notes from the Recursive event where monitoring is the number-one safety bet, plus the papers behind persona selection, emergent misalignment (”writing bad code makes you evil”), eval-gaming, and accidental chain-of-thought training.

Matthew Sanders (Longbeard / Magisterium AI) on Catholic AI after the Pope’s encyclical — what it was like at the Vatican, the divergence with Anthropic on machine consciousness, the red line on autonomous weapons, and why the last 5% of alignment is non-negotiable for a faith tradition.

We close on a live test of the cigarette-business refusal example from the OpenAI model spec, the tension between research and business “layers” in deployed models, and the argument that open-source AI may now be unbannable on religious-freedom grounds.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit briefing.ai-in-the-am.com

Trust and Recovery in AI. Andy Fernandez (HYCU) on backups as the enterprise black box, David Villalón & Manuel Romero (Maisa) on auditable digital workers, and Snehal Antani (Horizon3.ai) on the 77-second breach and why deception is the EMP of AI cyber weapons. Plus the disk trade and 1,766 miles of full self-driving.

(0:00) Opening — Why a daily AI show

(0:31) Why we need more AI shows

(11:12) What are we trying to do here

(17:23) Andy Fernandez (HYCU) — AI, cyber resilience & recoverability

(23:23) Backup data answers what was deleted

(27:14) Telemetry won't replace recovery

(43:14) Trust requires guardrails, not blind access

(48:21) David Villalón & Manuel Romero (Maisa) — Digital workers that survive production

(55:00) We don't use workflows, we handle exceptions

(57:50) Workflow thinking constrains knowledge work

(1:03:55) Compiling and executing programs on the fly

(1:09:58) ROI isn't just cost — it's faster value

(1:23:16) Snehal Antani (Horizon3.ai) — Trust through autonomous security validation

(1:25:50) Attackers chain small weaknesses

(1:47:49) Assume breach, manage blast radius

(1:58:08) Seventy-seven seconds to lose

(2:02:09) Deception is the EMP of AI

(2:03:02) Closing — The disk trade, Panopticon, and FSD

(2:05:46) Products get pulled by the market

(2:23:04) Crypto is banking's overlay

(2:26:18) Rules need to get better

Guests:

Andy Fernandez — HYCU (@hycuinc)

David Villalón — Maisa (@davipar)

Manuel Romero — Maisa (@mrm8488)

Snehal Antani — Horizon3.ai (@snehalantani)

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit briefing.ai-in-the-am.com