AI Inside

The Messy Web of AI Guardrails

AI Inside

Show Notes

This week Jason Howell and Jeff Jarvis unpack a surprising policy reversal from the White House, which is now weighing a plan to vet frontier AI models before release after revoking Biden's safety order on day one. They also dig into the Musk versus OpenAI trial testimony, where $80 billion Mars requests and near-physical confrontations made it feel more like a soap opera than a courtroom.Also in this episode: OpenAI traced a goblin obsession back to a rogue reward signal, Yann LeCun told students to ignore AI doom narratives, Gary Marcus called out Richard Dawkins for declaring Claude is conscious, a researcher invented a fake disease that every AI treated as real, and in the speed round, Nvidia's China market share hit zero, Anthropic launched a $1.5 billion enterprise joint venture, Google split its TPU into two chips, and China made it illegal to fire workers over AI. New episodes every Wednesday at aiinside.show.

Note: Time codes subject to change depending on dynamic ad insertion by the distributor.

CHAPTERS:

0:00 - Start

0:04:00 - White House Considers Vetting A.I. Models Before They Are Released

Behind the White House’s Potential Rethink on A.I.

0:13:51 - Canadian fiddler sues Google after AI Overview wrongly claimed he was a sex offender

What Was Discussed at Google’s White House Meeting About A.I.

0:22:32 - Top AI companies agree to work with Pentagon on secret data

0:24:09 - OpenAI’s president does ‘all the things,’ except answer a question

0:31:32 - Where the goblins came from (OpenAI)

0:36:29 - Video explaining JEPA with LeCun

0:42:50 - AI godfather Yann LeCun's blunt advice for the AI age

0:44:47 - Will A.I. Make College Obsolete?

0:48:51 - Marcus: Richard Dawkins and The Claude Delusion

0:55:29 - Scientists invented a fake disease. AI told people it was real

1:02:37 - Jensen says Nvidia now has 'zero percent' market share in China — says US export policy 'has already largely backfired'

1:04:40 - Anthropic Unveils AI Agents to Field Financial Services Tasks

1:06:12 - New: Higher usage limits for Claude and a compute deal with SpaceX

1:08:33 - Google's eighth generation TPUs: two chips for the agentic era

1:10:25 - China makes it illegal to fire humans if AI takes their jobs

Hosts: Jason Howell and Jeff Jarvis

Download and subscribe to AI Inside in audio and video: https://aiinside.show/

Support the podcast on Patreon for special perks: https://www.patreon.com/aiinsideshow. You'll get ad-free episodes, members-only Discord, T-shirts and stickers you love, and get ad-free audio and video feeds, a members-only Discord, and exclusive content. Learn more about your ad choices. Visit megaphone.fm/adchoices

More from AI Inside

Oct 22, 2026

This week Jason Howell and Jeff Jarvis dig into PHP2Shell, a pair of critical PHP bugs that sat in the code for more than a decade until AI helped find them, sending governments scrambling to pull websites offline. Then they look at people using Claude to make their smart TVs dumb again, synth fans vibe-coding new firmware onto old gear, and a researcher who says AI’s black box was never as tangled as we thought. Also in this episode: Meta’s Muse Spark Open, Nvidia’s DSX POD and OpenShell, DHS ordering DeepSeek out of federal systems, a Muse privacy leak, Disney’s Character.AI bankruptcy play, and Reflection’s Beam. New episodes every Wednesday at aiinside.show. Note: Time codes subject to change depending on dynamic ad insertion by the distributor. CHAPTERS: 00:00:00 - Show intro 00:12:46 - PHP2Shell: AI finds critical PHP bugs · Related 00:21:49 - AI agents strip ads from smart TVs 00:33:29 - Vibe-coded firmware for music gear 00:40:07 - Goodfire: neural networks may be more readable · Related · Research 00:50:26 - Meta opens Muse Spark · Related 00:54:09 - Nvidia DSX POD targets data-center efficiency 00:58:11 - Nvidia OpenShell for AI agents 00:59:42 - DHS orders DeepSeek removed from federal systems 01:02:43 - Meta Muse privacy exposure 01:04:19 - Muse downloads surge as ChatGPT slips 01:06:06 - Disney and Character.AI 01:07:21 - Reflection Beam 01:08:07 - Anthropic AI safety bill 01:10:17 - Bill Gross and AI-generated startup ideas 01:15:25 - Wrap-up and where to follow Hosts: Jason Howell and Jeff Jarvis Download and subscribe to AI Inside in audio and video: aiinside.show Support the podcast on Patreon for special perks. You’ll get ad-free audio and video feeds, a members-only Discord, T-shirts and stickers, and exclusive content.

Oct 15, 2026

This week Jason Howell and Jeff Jarvis ask whether we're witnessing the end of software, as developer Brandon Thomas uses Claude to build free, open-source clones of Microsoft Office and Adobe's Creative Suite while AI-decompiled versions of Halo, GTA, and Call of Duty run right in a browser tab. They also dig into OpenAI's release of 722 AI-written math papers and the backlash from mathematicians. Also in this episode: Jason shares his early access experience with Gemini 4 Argon, Meta launches its closed Muse Wonder models, the FBI arrests an OpenAI engineer accused of planting a hidden recording device, Anthropic warns about AI self-replicating on smart TVs, and Google Play opens up to personal vibe-coded apps. New episodes every Wednesday at aiinside.show. Note: Time codes subject to change depending on dynamic ad insertion by the distributor. CHAPTERS: 00:00:00 - Cold open and episode preview

Oct 8, 2026

This week Jason Howell and Jeff Jarvis dig into two big Western open-weight models that landed within 24 hours of each other. Mistral’s trillion-parameter Large 4 was trained entirely in Mistral’s own European data centers, and Reflection AI’s Apache-licensed Beam comes from a team of DeepMind veterans, with both aiming at the Chinese labs that have dominated open AI.

Oct 1, 2026

This week Jason Howell and Jeff Jarvis tackle AI agents from every angle. Nvidia released an open source safety platform just as OpenAI launched Dots, always-on agents with their own cloud computers and app integrations. Meta unveiled the Muse Charm, a palm-sized AI device. Anthropic's IPO prospectus revealed $518 billion in infrastructure commitments and extensive risk warnings. AMD announced an $8.2 billion acquisition of World Labs.

Sep 25, 2026

Thanks to Qualcomm for sponsoring Jason's travel to Maui for the Qualcomm Snapdragon Summit.

See all episodes →