The Rundown AI / Articles / AI / Anthropic's secret 'Mythos' model

Anthropic's secret 'Mythos' model

PLUS: Create Skills in ChatGPT with Codex

Zach Mink

March 30, 2026

Good morning, AI enthusiasts. OpenAI had Q* and Strawberry. Now Anthropic has its own 'accidental' preview of what's coming next.

Details of ‘Claude Mythos’ leaked via a CMS error, describing a system in a new tier above Opus with cyber capabilities Anthropic says are 'far ahead' of anything else available — in what looks like another step up the frontier ladder.

In today’s AI rundown:

Anthropic accidentally leaks ‘Mythos’ AI details
The Rundown Roundtable: Our AI use cases
Create Skills in ChatGPT with Codex
The personal war behind OpenAI and Anthropic
4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

ANTHROPIC

🔓 Anthropic accidentally leaks ‘Mythos’ AI details

Image source: @M1Astra on X

The Rundown: Details of Anthropic's next flagship AI, Claude Mythos, surfaced this week after the company's CMS left launch materials in an unsecured data store, with the leaked blog calling it 'a step change' and Anthropic’s most capable system to date.

The details:

A CMS configuration error left thousands of unpublished assets, including a draft blog post about the model, in a publicly accessible data cache.
The draft placed Mythos in a new “Capybara” tier that would sit above its Opus class, both larger and more expensive to run.
Anthropic flagged the model as “currently far ahead of any other AI model in cyber capabilities” and warned it could help hackers outpace defenders.
Anthropic confirmed to Fortune that a “new general purpose model with meaningful advances in reasoning, coding, and cybersecurity” is being tested.

Why it matters: A safety-focused AI lab 'accidentally' leaving its most powerful model's launch plans in a public data store has a similar vibe to OpenAI’s Q*-era leaks, where conveniently timed rumors doubled as free hype. Accidental or not, a new model tier above Opus sounds like another major next step up the frontier ladder.

TOGETHER WITH PINECONE

🧠 Easily add knowledge to your AI workflows

The Rundown: Pinecone Assistant is an end-to-end knowledge service that handles the heavy lifting behind AI retrieval like chunking, embedding, search, and reranking, so you don't have to build and maintain your own stack. Just upload your data and start querying.

With the new Pinecone Assistant n8n node, you can:

Turn any data source into knowledge for your AI app without building a retrieval pipeline
Upload files like PDFs, DOCX, JSON, TXT, and Markdown, and start chatting with your documents instantly
Connect sources like Google Drive, Slack, and webhooks for accurate, real-time answers

Start building for free.

THE RUNDOWN ROUNDTABLE

💡 The Rundown Roundtable: Our AI use cases

The Rundown: The Rundown Roundtable is a weekly feature where we poll members of The Rundown staff about how we use AI in our work and daily lives.

Nate, University Educator: I continue to use Claude artifacts for all kinds of visualizations. Recently, when purchasing acoustic panels for my ceiling, I wasn't sure how many to buy or the right orientation to install them in.

I had Claude create a mock-up of my room and then lay them out in different orientations and different patterns to optimize the right number, the right order, and the right way to install them. I made sure to purchase the right amount, reduced waste and cuts, and was able to better estimate costs when comparing different options.

I can see this being super valuable any time I'm doing any kind of home improvement project that includes estimating materials, whether it's tile, carpeting, or any kind of paneling across an area.

Shubham, Editor: Claude created my portfolio website in an instant. I prompted it with my links (LinkedIn and social), described what I wanted, and it built the entire thing in one sitting — design, deployment, DNS config, SEO, and Google indexing included. I made a handful of edits across the session, and it handled every single one without friction.

When things broke during deployment, it debugged in real time and fixed them via desktop extension browser control. Zero code written by me.

AI TRAINING

💪 Create Skills in ChatGPT with Codex

The Rundown: In this guide, you will learn how to get repeatable "skill"-style workflows to work with ChatGPT. The idea is to use Codex Desktop as a skills playground, create a reusable skill there, and then run it on demand.

Step-by-step:

Install the Codex desktop app, open it on your PC, click add a new project (CMD/CTRL + o), then create/select a folder to act as your skills playground
Start a thread and ask Codex to create the needed skill. Example: “Make me a skill to use the AARR framework to evaluate existing startups or business ideas”
If the skill does not appear right away, quit and reopen the Codex app. Then type / and select your new skill to run it like any other repeatable workflow

Pro tip: Type /skill creator to make sure Codex creates it the same way each time. You can also specify whether you want it to be a global skill or a skill just for this folder.

PRESENTED BY CDATA

🔬 Most MCP servers fail more than you think

The Rundown: 378 prompts across CRM, ERP, project management, and data warehouse platforms. Most MCP servers were accurate 60-75% of the time, silently dropping date filters, multi-condition queries half-applied, and write operations failing validation. But CData’s Connect AI achieved 98.5% accuracy.

CData’s report covers:

Where direct API translation fails and why
Common failure patterns by query complexity
How CData Connect AI achieved top scores

Download the full benchmark report.

DARIO AMODEI & SAM ALTMAN

⚔️ The personal war behind OpenAI and Anthropic

Image source: India AI Impact Summit

The Rundown: The WSJ just laid out the personal grudges, power struggles, and broken promises between Sam Altman and Dario Amodei that trace back to an SF group house in 2016, with the fallout shaping the rivalry between the two AI leaders.

The details:

Dario (2016-2020) and Daniela Amodei (2018-2020) worked at OAI prior to Anthropic, with the WSJ detailing early issues with co-founder Greg Brockman.
Brockman reportedly once floated selling AGI to UN Security Council nuclear powers, a proposal Dario considered 'tantamount to treason.'
The WSJ also reported that Altman accused the Amodeis of plotting against him to the board in a private meeting, then denied it when confronted.
Amodei privately likened Altman/Musk suit to Hitler vs. Stalin, called Brockman's pro-Trump PAC donation 'evil’, and compared OAI to Big Tobacco.

Why it matters: Kudos to the WSJ for these nuggets that paint a much deeper picture of the decade-long drama between Amodei and Altman. The grudges are entertaining, but they're also steering the trajectory of two of the most important AI companies — with impacts that ripple through much more than just a personal rivalry.

QUICK HITS

🛠️ Trending AI Tools

🧠 Adapt – The AI Computer for your company. It connects every tool, learns your business, and takes action for you directly in Slack*
🎨 Phota Studio - Phota's personalized photo editing and generation model
🧠 TRIBE v2 - AI that simulates brain responses to sights, sounds, language
🗣️ Cohere Transcribe - SOTA, open-source speech recognition model

*Sponsored Listing

📰 Everything else in AI today

Sam Altman reportedly told OAI staff he tried to "save" Anthropic during its Pentagon standoff, per Slack messages seen by Axios — even as OpenAI locked in its own deal.

xAI’s Ross Nordeen reportedly departed the company this week, who was the last remaining of the original 11 co-founders at the startup besides Elon Musk.

Pharma giant Eli Lilly entered a $2.75B deal with Hong Kong's Insilico Medicine to license its AI-discovered drug pipeline, with 28 compounds already in development.

Anthropic won a federal injunction blocking the Trump administration's supply-chain-risk designation, with the judge calling it "classic illegal First Amendment retaliation."

Google expanded the rollout of its Live Translate feature to iOS, turning any pair of headphones into a real-time interpreter across 70+ languages.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader David B. in Wichita, KS:

"My dad passed away in February, leaving behind a large box of handwritten letters from the 50's & 60's. I photographed them with my phone and fed the images in bulk to Claude Code — it read the cursive, transcribed everything, and helped me build a family history website with the original photos alongside each transcription.

Something that would have stayed in a box is now a family archive grandkids can actually explore."

How do you use AI? Tell us here.