🚨ANTHROPIC CEO: OPEN SOURCE AI IS GETTING DANGEROUS
Anthropic CEO Dario Amodei told lawmakers that open-source AI is moving down a “very dangerous path.”
His warns that once powerful models are released openly, companies lose the ability to monitor misuse, revoke access, or update safety guardrails.
停止されているFebleについてかなり急展開ですが、韓国で開かれたAnthropicのイベントで関係者が「数日以内に(in the coming days)使えるようになると確信している」と発言していたようで、意外と早くFableが戻ってくる可能性。
トップのアモデイがG7でトランプや閣僚と直接話した結果かもしれない
"The reason Fable 5 is unusable? Peak American politics, lol. You can make it perfectly safe, but if it doesn't align with what the government wants, suddenly you're getting sabotaged. Talk about shooting yourself in the foot. (Disclaimer: just my personal opinion here.)"
The Most Advanced AI Got My Name Wrong
Yesterday (June 9, 2026), Anthropic released Claude Fable 5.
It's the public version of the Mythos-class model. Because of cybersecurity concerns, it shipped with safety measures attached for general release. The benchmark numbers are reportedly the highest yet.
I tried it. A day late.
First, the honest part
It's solid.
The depth of reasoning may be above Opus. Long-form tasks, structural analysis, responses to critical questions—they all come back properly formed. "Impressive" isn't the wrong reaction.
But that alone doesn't say anything.
Without context, you only get textbook answers
I had it analyze the structure of my own project.
The information I gave it at first was thin. What came back was "as a first step, create one free session"—except the workshop is already scheduled and running. It started from a place I'd already passed.
Weak input, weak output. Obvious enough.
I updated the context to match reality. Opening date, the workshop announcement URL, the support I'm getting from the Chamber of Commerce, the structure of the competition—once I handed all that over, the quality of the responses changed.
Then it got my name wrong.
My own name. It was written correctly in the materials. It still came out altered. It's an unusual name, so something slipped somewhere in the conversion.
Whatever the cause, getting a name wrong in a business context is a non-starter. Whether it's someone else's name or your own, it's the same thing—what comes out has to be checked by a human, all the way to the end.
The numbers
Every time you send a question, the token usage shows up.
The actual figures: input 5,569 / output 721 tokens.
That's one question. On a subscription plan, Fable 5 counts as double usage. If you're going to use it continuously for real work, free or budget plans are out of the question. A MAX plan becomes the realistic baseline.
I'm on a Pro plan, and I'm a user who works mostly with Opus. This week I happened to hit 75% of my weekly limit by Tuesday. For someone who uses it that way, Fable 5's token consumption is a conversation that assumes a MAX plan.
The people who stop at "impressive," and the people who design
Every time a new product like Fable 5 drops, "you can do anything now" articles get mass-produced. That's nothing new.
But here's what this round of testing showed.
For someone who can design what to feed it and what to draw out, Fable 5 is genuinely a strong tool. For someone who touches it without designing anything, you get responses that look impressive—and don't hold up in actual work.
As an observation
Fable 5 can do situational assessment and structural analysis. Its responses to critical questions aren't off the mark, as long as the context is there.
But to draw out "a next step with its feet on the ground," you need materials built up from track record and numbers. That's not something that comes together overnight.
Whether real work can keep up with the pace of new releases—honestly, I still don't know.
I do think there's nothing to do but keep touching it.
This is a personal observation. Take it as a reference.
Araki Chosai / Resonance Project