Random Llama
Random Llama
ProductsSolutionsBlogCase StudyContact
Get a Quote
Weekly Newsletter

Get AI & productivity insights weekly

Privacy-first tools, workflow tips, and early product access. No spam — unsubscribe anytime.

Random Llama Software

The digital forge for privacy-first tools and high-performance web platforms.

Links
  • Home
  • Products
  • Case Study
  • Blog
  • Solutions
  • Credentials
  • Contact
Services
  • Custom CMS
  • Booking Engines
  • Mobile Apps
  • AI Integration
Connect
  • Privacy Policy
  • Terms of Service
  • Cookie Policy

© 2026 Random Llama Software, LLC. All rights reserved. Privacy Policy

Back to Blog
ai-toolsprivacyopen-source

Anthropic Leaks Its Own AI Model, OpenAI Kills Sora

Robert HattalaMarch 30, 2026

Anthropic Accidentally Revealed Its Most Powerful Model

Anthropic left nearly 3,000 internal files in a publicly accessible data store with no authentication. Among them: a draft blog post describing Claude Mythos, a new model tier above Opus that the company calls a "step change" in capabilities.

The leaked documents show Mythos outperforming Opus 4.6 across cybersecurity, coding, and academic reasoning benchmarks. Internally codenamed "Capybara," it sits in a brand new tier above everything Anthropic has shipped. The company is privately warning government officials that Mythos makes large-scale cyberattacks significantly more likely in 2026.

This is wild for two reasons. First, the security irony of an AI safety company leaving files unsecured is hard to ignore. Second, the cybersecurity warnings suggest these models are getting good enough at finding vulnerabilities that the offense-defense balance is shifting. If you build software, that means patching and security tooling just got more urgent.

OpenAI Killed Sora and Nobody Should Be Surprised

OpenAI announced on March 24 that Sora is done. The app shuts down April 26. The API gets six months, ending September 24. Disney found out less than an hour before the public announcement, despite having committed $1 billion to the partnership.

The numbers tell the story. Sora peaked at about a million users, then dropped below 500,000. It was burning roughly $1 million per day. Video generation at scale is just too expensive right now, and OpenAI decided to redirect that compute toward ChatGPT and their core API.

This matters for anyone building on AI APIs. Products can disappear fast when the economics don't work. OpenAI also quietly killed several other products this month. The lesson: build on capabilities you can replicate or self-host, not on a single provider's experimental feature.

Mistral Bets $830 Million on European AI Independence

France's Mistral secured $830 million in debt financing to buy 13,800 Nvidia chips and build a data center near Paris. The goal is 200 megawatts of compute capacity across Europe by end of 2027.

This is the clearest signal yet that Europe wants its own AI stack, not just models but actual owned infrastructure. For developers and small businesses outside the US, having a strong European provider with local data residency could matter a lot. Especially as AI regulation in Europe continues to tighten.

What This Week Tells Us

The AI industry is splitting into two tracks. One track is raw capability, with models like Mythos pushing into territory that worries even the people building them. The other track is economics, where even OpenAI can't make video generation profitable at scale. For builders, the practical move is the same as always: use what works today, don't depend on any single vendor, and take security seriously. The models are getting powerful enough that both the opportunities and the risks are very real.

Related posts

Anthropic's Mythos Leak and Mistral's $830M Bet on AI

Anthropic accidentally revealed its most powerful model yet, Mistral raised $830M for European AI infrastructure, and AI lobbyists are spending big on midterms.

March 30, 2026

Mistral Grabs $830 Million for AI Chips and Europe Fights Back

Mistral secured $830M in debt to buy 13,800 Nvidia GPUs. A $100M political fund backs pro-AI candidates. AWS trains 100K people for free. The money is loud.

March 30, 2026

ARC-AGI-3 Humbles Every AI Model While Arm Ships Its First Chip

ARC-AGI-3 scored every frontier model under 1% on tasks humans ace. Meanwhile Arm built its first chip in 35 years and Google went live in 200+ countries.

March 29, 2026
All posts