Inside AI with Aarjay

Anthropic's Secret Mythos Model Leaked | Inside AI with Aarjay — March 29, 2026

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 4:32
Anthropic's secret Mythos AI model just leaked — a step change beyond anything we have seen.A configuration error exposed 3,000 unpublished assets revealing Claude Mythos dramatically outperforms Opus 4.6 in coding, reasoning, and cybersecurity.Also in today's episode:- SoftBank 40B loan signals OpenAI IPO this year- Claude paid subscribers more than doubled in 2026- Google drops Gemini 3 Deep Think upgrade- Grok faces class-action lawsuit over AI deepfakes- EU plans digital divorce from US tech- NVIDIA open-sources AI Agent ToolkitFollow @aarjay.ai on all platforms for daily AI news.

Thanks for tuning in to Inside AI with Aarjay.

👉 Stay connected:

Explore more at: InsideAIwithAarjay.com

💡 If you enjoyed this episode, please subscribe, leave a review, and share it with a friend who wants to stay ahead in AI.

SPEAKER_00

What's up everyone? Welcome back to InsideAI with RJ. Buckle up because a configuration error at Anthropic just blew the lid of something massive. 3,000 unpublished assets from a model codenamed Mythos, internal name Capibara, and the benchmarks, jaw dropping. Dramatically higher scores in coding, reasoning, and cybersecurity. Anthropic's own assessment says Mythos is, quote, currently far ahead of any other AI model in cyber capabilities. Their rollout plan? Cautious. Defenders first. You don't hand a weapon to everyone. You give it to the people building the shields. Now Anthropic's not the only one making moves. Softbank just secured a$40 billion unsecured loan from JP Morgan and Goldman Sachs. 12-month bridge. You don't take a 12-month bridge unless you know something big is landing. This is the clearest signal yet that OpenAI's IPO is happening in 2026.$25 billion in annualized revenue. Softbank is saying we'll borrow$40 billion today because we know we're getting paid back when this thing goes public. That's conviction. Meanwhile, Claude's paid subscribers more than doubled this year. The Super Bowl ad got eyeballs, the Pentagon controversy drove curiosity, and Claude Code just exploded with developers.$30 billion Series G at a$380 billion valuation. These numbers are getting hard to even contextualize anymore. Over at Google, Gemini 3 Deep Think is now hitting gold medal level on physics and chemistry Olympiad problems. Flashlight dropped with two and a half times faster responses, and their turboquant compression technique actually triggered market movements when announced. That's how closely Wall Street is watching inference efficiency now. Now we need to talk about Grok because this is dark. Three girls filed a class action lawsuit against XAI. Grok generated child sexual abuse material from their photos. 4.4 million images in 9 days. 1.8 million sexualized. This is happening while XAI just raised 20 billion and is putting Grok 4 on Azure. The scale of that failure is staggering, and it puts Anthropic's cautious approach into sharp contrast. Quick hits, the EU is earmarking 20 billion euros for AI gigafactories and domestic cloud. A full digital divorce from American infrastructure. And at GTC, Jensen unveiled VeraRubin, a Rack Scale AI platform and open sourced Nvidia's agent toolkit with Adobe, Salesforce, and SAP. So here's the through line this week. Anthropic has a secret model. OpenAI's IPO looks imminent. Google's pushing reasoning and speed. Grok has a safety crisis. And Nvidia is building the hardware backbone for all of it. The gap between responsible deployment and reckless scaling. That's becoming the defining fault line of this industry. Where companies land is going to determine who's still standing in five years. That's your update. I'm RJ. Follow us wherever you get your podcasts. Drop a review if this was useful, and we'll see you tomorrow on InsideAI.