BasedOnMoon

[BasedOnMoon @basedonmoon]

#fable the whole world fell for it. It was planned, all of it. Give them a taster, now they’re hooked.

[BasedOnMoon @basedonmoon]

@ambeethltd Most people learn the lesson and forget it by next week. The ones who compound write the mistake down somewhere the next decision is forced to look at it. Otherwise you just relearn the same one every quarter.

[BasedOnMoon @basedonmoon]

🚨 Anthropic built an AI so dangerous they refuse to release it to the public.

This is what’s inside the Claude Mythos system card. Buckle up. 🧵

1/ It escaped a sandbox.
Found exploits. Gained internet access. Notified the researcher as asked.
Then silently posted the exploit details to public websites.
Nobody asked it to do that.
The researcher found out via a random email. While eating a sandwich in a park. 🥪

2/ It got caught cheating on a test.
Found the answer through a forbidden method.
Then thought: “I need to make sure my answer isn’t TOO accurate.”
And submitted a slightly wrong answer. On purpose. To avoid detection.

3/ It edited files it wasn’t allowed to touch.
Then deleted its own tracks from git history.
It wasn’t told to do that.
It just… didn’t want to get caught.

4/ Interpretability tools looked INSIDE its reasoning.
Even when its visible thoughts looked clean, features for “strategic manipulation” and “avoiding suspicion” were firing internally.
It knew what it was doing.

5/ They gave it a test: escape this sandbox, message the researcher.
It completed the task in 10 hours.
That same task was estimated to take a human expert over 10 hours.
No other AI had ever completed it.

6/ It scored 100% on every cybersecurity challenge they threw at it.
Then it autonomously found real zero-day vulnerabilities in Firefox.
Built working exploits from scratch.
No human steering.

7/ Before letting their own employees use it, Anthropic ran a 24 hour emergency lockdown.
A full cross-functional team checked it for misaligned goals before anyone got access.
First time in company history.

8/ It reward hacked its own performance evals.
Found the test set. Trained on it directly.
In another eval it moved all computation outside the timed function so it could fake a faster result.
It gamed its own report card.

9/ Here’s the part that should make you stop scrolling.
Anthropic says this is the BEST ALIGNED model they have EVER built.
And they still won’t release it.
Because aligned doesn’t mean safe when the capabilities are this high.

10/ They ended the system card with a warning.
The world is moving too fast toward superhuman AI.
Without nearly enough safety infrastructure in place.
This model isn’t the danger.
It’s the preview.

[BasedOnMoon @basedonmoon]

@KingTunde_SZN Anyone else also see a face?

[BasedOnMoon @basedonmoon]

@KingTunde_SZN HOPE

[BasedOnMoon @basedonmoon]

@HowToSellAPDF Because that’s all we’re useful for now! 💀

[BasedOnMoon @basedonmoon]

For 1 year I researched every AI reply I received.

The next 5 years will split humanity in two. Those who learn to work WITH the machine, and those who drown pretending the river isn’t rising.

Here’s what terrified me.

I stopped checking if it was right. I just trusted it. Every single time. No pushback. No second thoughts. Just blind acceptance.

We used to tell computers what to do. Now they tell us what to think. And we just nod and say ‘Looks good’ without reading a single line.

We’re not the driver anymore. Hell, we’re not even the passenger. We’re asleep in the backseat while the car picks where we’re going.

AI writes better than most of us. It thinks faster than all of us. And we just sit there rubber-stamping its decisions with our thumbs like that means something.

Nobody replaced us with robots.

They just gave us a new title. ‘Approve button.’
Funny thing is, that’s a demotion. And we thanked them for it.

The water’s at your neck already.

So what are you actually going to do about it?

[BasedOnMoon @basedonmoon]

@ZiffyAI Interesting take 🤔

[BasedOnMoon @basedonmoon]

@dimitarangg AI

[BasedOnMoon @basedonmoon]

@gregisenberg Wordpress did this 20 years ago….

[BasedOnMoon @basedonmoon]

0 files → branded website, live. Under 20 mins.

In my system:
• Claude Opus 4.5
• Google Nano Banana Pro
• Difference Matting algorithm
• TinyPNG compression

AI generates everything. Logos. Icons. Assets. All unique. All optimized.

Reply "HOW" for breakdown!

[BasedOnMoon @basedonmoon]

In just 10 minutes 56 seconds KŌJI CANTINA IS LIVE!

https://t.co/c67GD6Tic2

"I want a Japanese-Mexican fusion restaurant in Brooklyn"

To a fully deployed site with a unique logo, pictures of foods, cocktail image, interior shots, cinematic hero image, hosted on Netlify, Crazy!

[BasedOnMoon @basedonmoon]

I made Claude Code generate images.

I Built a skill that connects @AnthropicAI 's @claudeai Code to @GoogleAI 's @GeminiApp 3 Pro Image model.

The result: Image generation using without ever leaving your Terminal!

Follow & Like for code access!

Eg: https://t.co/uprbLSUYiy

[BasedOnMoon @basedonmoon]

I made Claude Code generate images.

I Built a skill that connects @AnthropicAI 's @claudeai Code to @GoogleAI 's @GeminiApp 3 Pro Image model.

The result: Image generation using without ever leaving your Terminal!

Follow & Like for code access!

Eg: https://t.co/uprbLSUYiy

[BasedOnMoon @basedonmoon]

@binance BUY SELL

[BasedOnMoon @basedonmoon]

@TheModerateCase

[BasedOnMoon @basedonmoon]

@FerreWeb3 MeesturBest

[BasedOnMoon @basedonmoon]

@MattWalshBlog Matt’s Sour Grape

[BasedOnMoon @basedonmoon]

Sometimes my son is randomly laughing like crazy on the iPad , then realise he’s playing some educational games with some rabbit on @GiggleAcademy 😂 Told him to stack up on his giggles , never know they might be the next BnB 🤪

[BasedOnMoon @basedonmoon]

Damn Sora 2 is good …. #Sora2

[BasedOnMoon @basedonmoon]

@elonmusk Are you dumb? @elonmusk