Anthropic's secret 'Mythos' model
Anthropic accidentally leaked details about its upcoming 'Claude Mythos' AI model through a CMS error, revealing it will be in a new 'Capybara' tier above Opus with advanced cyber capabilities. The leak included draft blog materials describing it as 'far ahead' of other AI models in cybersecurity, though Anthropic confirmed they are testing a new model with advances in reasoning, coding, and cybersecurity.
Summary
Anthropic experienced a significant data leak when a CMS configuration error left thousands of unpublished assets in a publicly accessible data cache, including draft materials about their upcoming flagship AI model 'Claude Mythos.' The leaked blog post described Mythos as 'a step change' and Anthropic's most capable system to date, positioned in a new 'Capybara' tier that sits above the current Opus class and is both larger and more expensive to run. The company 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 they are testing 'a new general purpose model with meaningful advances in reasoning, coding, and cybersecurity.' The newsletter also covered other AI industry news, including personal rivalries between AI leaders Sam Altman and Dario Amodei dating back to 2016, various AI tool updates, and reader examples of AI applications. Staff members shared their AI use cases, including using Claude for home improvement visualizations and website creation, while readers described using AI for family history projects involving transcribing handwritten letters.
About this episode
PLUS: Create Skills in ChatGPT with Codex
Key Insights
- Anthropic accidentally revealed their most advanced AI model 'Claude Mythos' through a CMS error that left launch materials in a publicly accessible data store
- The leaked materials positioned Mythos in a new 'Capybara' tier above Opus, describing it as 'far ahead of any other AI model in cyber capabilities' and warning it could help hackers
- The WSJ reported that personal grudges between Sam Altman and Dario Amodei trace back to 2016, with Amodei reportedly comparing the Altman/Musk situation to 'Hitler vs. Stalin' and calling Brockman's pro-Trump donation 'evil'
- Dario Amodei considered Greg Brockman's proposal to sell AGI to UN Security Council nuclear powers as 'tantamount to treason' during their time working together
- The author suggests the 'accidental' leak has similarities to OpenAI's Q*-era leaks, implying these convenient rumors may serve as free publicity for major AI developments
Topics
Transcript
Good morning, {{ first_name | 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. 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 ANTHROPIC Image source: @M1Astra on X The Rundown: Details of Anthropic's next flagship AI, Claude Mythos, surfaced this week after the…
Full transcript available for MurmurCast members
Sign Up to AccessMore from The Rundown AI
Economists, researchers put AI’s job shock on the clock
Over 200 AI researchers and Nobel laureates signed a Stanford statement warning that AI could displace jobs at historic scale within the next decade, requiring immediate government action on safety nets and labor policy. Meanwhile, the AI industry continues to evolve with new tools, research findings on AI personality variations, and ongoing feuds between major figures like Musk and Altman.
Apple takes OpenAI to court
Apple has filed a lawsuit against OpenAI alleging the company poached over 400 Apple employees and used them to steal confidential hardware secrets for an unreleased device designed by Jony Ive. The newsletter also covers new AI tools, training resources, and research on long-term AI scenarios.
OpenAI sends GPT-5.6 to Work
OpenAI launched GPT-5.6 with three tiers (Sol, Terra, Luna), featuring near-Fable performance at lower costs and introducing ChatGPT Work with desktop integration. The newsletter also covers Meta's Muse Spark 1.1 release and argues that AI will concentrate demand among top 1-5% professionals while displacing mid-tier service providers.
SpaceXAI, Cursor release the strongest Grok yet
SpaceXAI and Cursor released Grok 4.5, a powerful new AI model that combines strong performance with faster speeds and lower costs than competitors like Claude and GPT. The newsletter also covers OpenAI's new GPT-Live voice model, ByteDance's Seedream 5.0 Pro image generation tool, and several other AI developments in the industry.
Meta climbs the AI image leaderboard
Meta released Muse Image, an in-house AI image model ranking No. 2 on leaderboards behind only OpenAI's GPT Image 2, with a teased video model also performing strongly. Meanwhile, Beijing is considering restrictions on Chinese AI model exports, creating potential geopolitical reciprocity risks similar to U.S. export controls, while companies like DoorDash demonstrate that pairing multiple AI models significantly improves code review accuracy and cost-effectiveness.