Good morning {{first_name| Artificial intelligence lovers}}. Anthropic said that Mythos was too dangerous to be released to the public. Weeks later, a Discord group was reported to be using it daily.
Working around naming conventions leaked in the data breach and loan contractor logins, the group has slid into Anthropic’s most restrictive AI yet – a bad start for keeping increasingly powerful systems under wraps.
-
Closed anthropic myths leaked out
-
SpaceX is investing $60 billion in AI programming startup Cursor
-
Use this dictation strategy to write better documents
-
Codex-powered ChatGPT agents for teams
-
4 new tools for AI, community workflows, and more
Anthropic
Image source: Images 2.0 / Rundown
Rundown: It is said to access Anthropic’s Mythos model It leaked in a Discord group within days of launch, after users guessed the company’s post and naming URL using patterns leaked in the recent Mercor hack.
-
It was a cybersecurity model Released On April 10 to select partners under “Project Glasswing”, as Anthropic deemed it too strong to be released to the public.
-
A private Discord group that tracks unpublished models accessed Mythos on release day and was using it regularly, Bloomberg reported.
-
One member had vendor credentials through contract work, with leaked Mercor details helping the group locate and access Mythos online.
-
The group told Bloomberg that it does not use Mythos for cyberattacks or other malicious activity, and also claimed access to other unreleased models.
Why it matters: The first alleged unauthorized use of the AI model that prompted the White House and others to call emergency meetings did not come from China, Russia, or any other competing nation, but rather from a random Discord group. This is not a great start, and the problem only gets worse as partners become more accessible and models become more risky.
Along with Bland
Rundown: Meet Norm – Bland’s new AI assistant that automatically builds production-ready audio agents. Simply describe what you want the agent to do, and Norm will take care of the routing, logic, and setup for you.
-
Rotate voice agents to schedule appointments, support customers, or qualify leads
-
Build safely in an isolated branch without touching your live production agents
-
Run agent-on-agent simulations to test performance before going live
SpaceX and the indicator
Rundown: SpaceX only Announce A new partnership with programming startup Cursor and an option to acquire the company for $60 billion later this year gives Elon Musk a shortcut in a race in which xAI spent the year losing ground to Anthropic and OpenAI.
-
CEO Michael Troll He said Each version of Cursor’s Composer model has reached a computational ceiling, with SpaceX’s Colossus now providing the required power.
-
Cursor is guaranteed $10 billion for the partnership, regardless, with the full buyout worth $60 billion only if Musk exercises it before the end of the year.
-
xAI boiled Courseor is led by Andrew Milich and Jason Ginsberg last month, with Musk saying the startup was “not built right the first time.”
-
It was the indicator Set to raise $2 billion at a $50 billion valuation before the deal with SpaceX Postponement regarding the acquisition due to the complexities of the IPO process.
Why it matters: Musk has tried and so far failed to build a leading cryptographic tool within xAI, with Grok having no answer to Claude Code or Codex. The deal swaps in-house build with an outside producer to scale up quickly, and gives Cursor the ability to potentially transform into a SpaceX-powered coding lab rather than a compute-hungry startup.
Artificial intelligence training
Rundown: In this guide, you’ll learn a two-step phonetic dictation system that makes you write better instantly and improves itself over time.
-
Steady No type To dictate, open a Codex/Claude Code session, and dictate: “Draft an outline for a short internal memo about [topic]”
-
Here’s the main part. Tell the agent: “Save this as a rough draft and do not edit it. Now create a separate working draft that we can review.”
-
Read it and use Typeless to add comments, point out things that are inaccurate, then things that are missing, and finally statements that seem generic or too AI
-
Prompt: “Rewrite the draft using the comments you left. Write it in my style. Use verbiage. No dashes. Keep the main points, but cut out anything that seems generic.” Review the rewrite and make another pass and edit if necessary
Pro Tip: Have the agent compare the untouched rough draft to the final drafts each week. It will extract the edits you make and update the editing rules for the future.
Provided by Braintrust
Rundown: Most teams charge AI applications with emotions. Braintrust’s free evaluation course fixes that by walking you through everything from writing your first evaluation to enrolling multi-step agents in production.
In this course you will learn:
-
How to use LLMs to automatically record results
-
Build test cases from real production failures
-
Analyze agent workflow to detect failure patterns
Obinay
Rundown: OpenAI only foot ChatGPT Workspace Agents, new Codex-powered cobots designed to handle multi-step team workflows autonomously across ChatGPT and Slack.
-
Workspace agents are shown as an “evolution” of single-user GPTs for 2023, with legacy GPTs still working for now and a conversion tool coming soon.
-
Powered by Codex in the cloud, agents can reserve memory, connect to connected applications, live in Slack or run on a schedule when users are offline.
-
Within OAI, salespeople use this feature for account research and follow-up drafts, while accounting turns it on for journal entries and reconciliations.
-
Custom agents can be created via ChatGPT and shared across teams, with the ability to set limits on data usage, approvals, and permissions.
Why it matters: OAI’s enterprise push is no secret, as workspace agents are solving a real problem – each team has accumulated scattered claims and patchy workflows over the past two years, and only a few have been able to unify them. The initial GPT store did not last, but agent upgrades and a corporate shift could help this debut find a better fit.
Anthropic Face it Backlash after Claude Code removed some new users in the Pro tier, with the company saying it was conducting a “small test” of the subscription flow.
Google unveil The new 8th generation TPUs are designed specifically for customer workloads, separating training and inference into two separate chips for the first time.
ideogram Fired Custom templates, allowing users to fine-tune image generation on 15-100 of their assets for consistent, brand-specific output.
Google open 75% of its internal code is now generated by AI, as the company has made significant gains in security and operations through AI and agent applications.
Odyssey foot Odyssey-2 Max, a global model three times larger, topped benchmark scores for real-time physics and is now in private beta.
Alibaba Coin Team Open source Qwen3.6-27B, the 27B model surpasses its predecessor 397B via higher coding standards.
In each newsletter we show how the reader is using AI to work smarter, save time or make life easier.
Today’s workflow comes from the reader Brigid H. In New Zealand:
“I created a complete app for our Claude farming business to record and track all stock purchases, sales, deaths and health treatments. The app has a full section for all invoices and any other files and attachments. This will track any stock purchases and treatments such as cutting and dipping. It has been a game changer!”
How do you use artificial intelligence? Tell us here.
That’s all for today!
Before you go, we’d love to know what you thought of today’s newsletter to help us improve The Rundown experience for you.
Rowan, Joey, Zack, Shubham and Jennifer – the humans behind The Rundown
