I have some really wonderful news for you today. Anthropic, one of the world's leading AI developers, has decided to protect us.
And I mean that sincerely.
What Project Glasswing actually is
Anthropic recognized that what they've developed could have catastrophic consequences in the wrong hands. By wrong hands they mean, of course, you and me. So they decided to put it in the right hands. Namely American mega-corporations - including Google, Microsoft, Amazon, and Apple - that answer to the American president. The model is called Claude Mythos Preview. It finds thousands of zero-day security vulnerabilities. And we don't get it.
The new two-tier society for AI
So the decision has been made to give the most intelligent AI model ever developed exclusively to selected large corporations. Not to the public. Not to developers. Not to you.
It's a bit like someone on a first date saying: I find you so attractive that I never want to see you again. Luckily, Google and Microsoft are getting it instead. Those two. The companies that have proven time and again over the past years that their only interest is the well-being of their users - and certainly not the stock price.
Claude Mythos Preview: too dangerous for you
The model autonomously finds thousands of security vulnerabilities in software - zero-day flaws that were previously unknown. Anthropic has decided that this tool may only be deployed in government-regulated environments and with selected enterprise partners. The reasoning: in the wrong hands, it could become the most powerful cyberweapon in history.
Your quota disappears faster now
If you feel like your AI subscription is delivering less than it did in March - you're right. Anthropic has adjusted how quotas are calculated during peak hours. Claude Code and agent-based workflows consume significantly more tokens than simple chat requests. This means Max subscribers can burn through their hourly quota in minutes. Anthropic has officially confirmed this.
Reduced thinking depth - without announcement
At the same time, the so-called "Extended Thinking" depth of the models has been adjusted - quietly and without a major changelog. If you want more thinking, you need more tokens. If you need more tokens, you hit the limit faster. That's called exclusivity. Or as the video puts it: you pay more for less.
We know this pattern already
Intelligence gets more expensive, limits get lower, and you're still supposed to keep using it enthusiastically. It's not about what you personally get out of it. It's about the fact that the big companies of this world are protecting you from your own stupidity. That's what they do. Always. For your own good. And coincidentally they also happen to make money doing it.
What this means
For my part, I'm perfectly happy to pay more for less - knowing that the really dangerous stuff is in safe hands with the right people. They're looking out for us. As they always have.
And that, ladies and gentlemen, is somehow enough.
The most important takeaway
The most dangerous AI in the world belongs to the corporations that can afford it. Everyone else pays more for less - and is supposed to understand that as a protective measure. I'll leave the conclusion to you.
Sources
-
Anthropic: Project Glasswing (official)
Anthropic's initiative for AI-assisted cybersecurity with Claude Mythos Preview - available only to selected partners:
anthropic.com - Project Glasswing -
VentureBeat: Anthropic says its most powerful AI cyber model is too dangerous to release publicly
Reports on the decision not to make Claude Mythos Preview publicly available:
venturebeat.com - too dangerous to release -
The Hacker News: Anthropic's Claude Mythos Finds Thousands of Zero-Day Flaws
Documents the capabilities of Claude Mythos Preview and the restricted access:
thehackernews.com - Claude Mythos Zero-Day -
PCWorld: Anthropic confirms it's been 'adjusting' Claude usage limits
Anthropic officially confirms the adjustment of usage limits for Claude subscribers:
pcworld.com - usage limits adjusted -
The Register: Anthropic tweaks Claude usage limits to manage capacity
Reports on the background of quota adjustments during peak hours:
theregister.com - tweaks usage limits -
DevClass: Anthropic admits Claude Code users hitting usage limits 'way faster than expected'
Anthropic engineer confirms: users are hitting limits significantly faster than anticipated:
devclass.com - usage limits faster than expected