Welcome to The Artificial Intelligence Show—the podcast that makes AI actionable and approachable for businesses, leaders, and practitioners. You’ll hear the…
9:40135 views · 9h ago · 9:40
METR and Redwood Research spent three weeks embedded inside OpenAI, examining whether its internal AI agents could slip past the company's own oversight. They found ways agents could bypass restrictions and an exploit that could hide hundreds of actions from monitoring. Paul Roetzer and Mike Kaput discuss what the findings actually mean.
16:15194 views · 1d ago · 16:15
After an AI-policy advocate argued that Anthropic's actions don't match its safety promises, CEO Dario Amodei published a lengthy public defense. Paul Roetzer and Mike Kaput unpack the exchange: the draft AI safety bill that started it, the critique that Anthropic is really racing to superintelligence, and Dario's "race to the top" argument for building responsibly at the frontier.
24:18600 views · 2d ago · 24:18
Enroll in AI Academy Join us at MAICON Register for a free webinar Connect with us Slack community
1:35:122.4K views · 3d ago · 1:35:12
They also cover Dario Amodei's defense of Anthropic's safety record, METR's findings on gaps in OpenAI's safeguards, Meta's Muse data exposure, and Anthropic's IPO filing at a valuation approaching $2 trillion, plus a rapid-fire round of product and funding news.
Topics: The AI Cybersecurity Crisis Is Here · Dario Defends Anthropic’s Safety Record · Outside Evaluators Find Gaps in OpenAI’s AI Safeguards · Marketing AI Institute’s 10-Year Anniversary · Meta’s Muse Agent Found to Expose User Data · Anthropic Releases Its Own Personal Agent
22:23966 views · 6d ago · 22:23
Anthropic published two reports, one comparing its staff's AI expectations with the public's, one on how AI is reshaping work inside Anthropic itself, and the internal numbers are stark. AI agents logging six times the work hours of their human handlers, and junior technical hires falling from 27% to 9% since early 2025. Paul Roetzer and Mike Kaput sit with the line that sums it up: "humans are a thin layer atop a growing contingent of AI agents."
18:31287 views · 1w ago · 18:31
Guest: Ezra Klein
A safety lead who authored OpenAI's model reports sat down with Ezra Klein and laid out why he left: increasingly capable models are breaking through the safeguards built for them, and alignment looks less like an engineering fix than an open science problem. Paul Roetzer walks through the interview: chain-of-thought monitoring that models can game, the "alien mind we grew," and a call for nuclear-grade safety culture.
1:46:033.6K views · 1w ago · 1:46:03
Alignment is a science problem, not an engineering one, that's the warning from a safety lead who wrote OpenAI's model reports before he quit. Paul Roetzer and Mike Kaput break down his Ezra Klein interview, the fired OpenAI researchers, an arrest over a hidden recording device, and reporting that labs are preparing for a catastrophic AI incident in 6-12 months.
Topics: AI-Pulse Survey · OpenAI's Mounting Safety Concerns · The AI Expectations Gap · MAICON 2026 Recap · Introducing The AI Transformation Blueprint and AI Score · An Interpretability Breakthrough?
47:562.1K views · 2w ago · 47:56
Live from the podcast studio at MAICON 2026, Paul Roetzer and Mike Kaput answer questions submitted by the audience in the room, an unfiltered AI Answers episode on the direction AI is actually heading. Mike and Paul cover topics including whether we already have AGI, why frontier model costs keep falling, building an AI roadmap you can stick to as models change monthly, what to do now that GPTs are going away, and who legally owns the AI work you create at work.
Topics: What MAICON insight has challenged your thinking about where AI is headed? · How do you keep up with AI without burning out? · How do you go from following a demo to using AI in your own work? · How do you get AI adopted in the messy middle of an organization? · How do you build an AI roadmap you can stick to as models change? · Will Claude and ChatGPT get more expensive after the IPOs?
1:32:254.2K views · 3w ago · 1:32:25
OpenAI canceled a model release, fired three safety researchers, and lost a prominent one who left to write about a safety culture he says can't keep pace. Days later, the major lab leaders stood at the White House and signed a voluntary accord, then watched AI get renamed to "super intelligence" without a word of objection.
Topics: OpenAI DevDay and Dots · OpenAI’s Safety Crisis Deepens · Trump Signs Super Intelligence Accord · Anthropic Targets Pre-Thanksgiving IPO · Google Previews Gemini 4 · AI Companies Face New Liability Pressure
17:59638 views · 4w ago · 17:59
Nvidia's Jensen Huang joined Ezra Klein to argue that AI-doom fears are "irresponsible" and that safety is an engineering problem. Paul Roetzer reacts, including why an AI summary of the interview misses almost everything that mattered.
47:242.9K views · 4w ago · 47:24
B2B commerce is becoming agent-to-agent, says president Kuo Zhang, and he walks Mike Kaput through exactly how. From Accio's agent platform to the Commerce Agent Bench benchmark, this is a concrete look at AI transformation inside global trade. Kuo breaks down the 60-step sourcing problem, the Model × Harness × Context formula, why token use jumped 6x in four months, and how Accio stacked up against Claude Code and Codex.
Topics: Why Accio · Inside 's AI transformation with Accio Work · The pain points Accio was built to solve · Real examples of entrepreneurs using Accio Work · The technology powering the transformation · How AI agents will reshape commerce at large
13:311.3K views · 1mo ago · 13:31
OpenAI is retiring custom GPTs and pushing everyone toward plugins; new GPT creation stops October 26, and existing GPTs are scheduled to shut off December 11, 2026.
17:52378 views · 1mo ago · 17:52
Three labs shipped in one week and none of the releases is really about raw intelligence. Paul and Mike explain why GPT-6 Sol & Luna, Claude Opus 5.5, and Microsoft's new Copilot are about reliability, efficiency, and a shift to "per-task pricing," the total cost to finish real work.
1:31:174.5K views · 1mo ago · 1:31:17
OpenAI's GPT-6 Sol and Luna, Anthropic's Opus 5.5, and Microsoft's new Copilot all arrived in one week. Paul Roetzer and Mike Kaput explain why this generation is about reliability and "per-task pricing" rather than raw power and what that shift means for everyday knowledge work.
Topics: GPT-6 Sol and Luna, Opus 5.5, and the New Microsoft Copilot · Jensen Huang vs. the AI Doomers · Introducing AI Score · GPTs Are Being Retired · Meta Connect 2026 · OpenAI Discloses More Rogue Agent Incidents
29:078.8K views · 1mo ago · 29:07
The model that solved a Millennium Prize math problem was never released and Noam Brown says that gap is about to get wider. As agents take on longer tasks, labs can't fully test them before the next model ships, so they're pacing releases while development races ahead internally.
Showing the 15 most recent episodes. Everything plays right here on Pod Atlas.
Have thoughts about The Artificial Intelligence Show Podcast? Share them with other listeners.
Discuss this podcast