Claude just found 1,244 photos of my 19y/o sister on a random server
She never posted a single one of them…
I was making a birthday post for her and reverse-searched an old photo of us from last Christmas
Google found it on a site I'd never heard of. Just an open folder, no password
Inside the folder - 4 subfolders:
/gallery - 2,312 photos, sorted by date
/screens - 1,140 screenshots of her bank app and WhatsApp
/docs - her ID card, front and back
/new - empty
I refreshed the page
/new - 1 file
A selfie. Taken 4 minutes ago
I called her. She had no idea
I asked if I could check her phone. She said yes
I plugged her Samsung into my laptop, turned on USB debugging and gave Claude Opus 5.5 one prompt:
"Find whatever is sending this phone's photos to the internet"
It scanned 214 apps in 90 seconds and stopped on 1
"System Sync"
No icon. Hidden from the app drawer. Not from the Play Store
Every permission it had, and what it used it for:
accessibility → to read every screen she opened
all photos → to copy her whole gallery
mic + SMS + location → to know where she was and who she talked to
device admin → so the "Uninstall" button stayed grey
Then Opus decompiled the app and rebuilt its history:
Day 1 - installed by hand, on her unlocked phone
Day 2 - full gallery uploaded, 1,244 photos
Day 3 - first screenshot of her bank app
Day 41 - 1.8 GB sent, mostly between 1am and 5am
And one line in the code - the server address
The same site where I found her photos
96 screenshots and 87 new photos were still waiting in the upload queue
What we did in the next hour:
1. Screen-recorded everything as evidence
2. Turned off device admin, then removed the app
3. Changed 14 passwords from my laptop, not her phone
4. Turned on 2FA everywhere
5. Reported the folder to the hosting company
2 days later the folder was gone
But someone held her unlocked phone for a few minutes, 41 days ago
She still doesn't know who
Your must have 3-minute check tonight after saving this post:
1. Battery - anything draining power with the screen off
2. Photos - only your camera and gallery need "all photos"
3. Accessibility - every app there needs a reason to be there
4. Device admin - most people only need "Find My Device"
5. Hidden apps - no icon + "system", "sync" or "health" in the name = check it
Found something? Don't delete it on the spot. Screenshot everything, then remove it and change your passwords from a different device
Prompt below ↓
Readers added context they thought people might want to know
The post claims a real incident where Claude scanned a Samsung phone via USB debugging to find stalkerware, but the video shows iPhone UI and a custom non-Claude interface, contradicting Android-only ADB and actual Claude tool integration limits.
x.com/MahdiSPHP/stat…
androidauthority.com/how-to-use-adb…
phonebox.dev/guides/how-to-…
Here's the prompt. Plug your Android into your laptop, turn on USB debugging, open Claude Code and paste this:
You are a phone security auditor. An Android phone is connected via adb. This is my own phone (or I have the owner's permission). Do NOT delete, disable or change anything - read only.
Step 1 - Check that adb sees the device. If not, tell me how to fix it.
Step 2 - Scan every app and collect:
- package name, app name, install date, installer (Play Store or installed by hand)
- apps with no launcher icon
- apps with "system", "sync", "health", "service" or "update" in the name that are NOT real system apps
Step 3 - Check the risky access:
- accessibility services that are turned on
- device admin apps
- apps with access to photos/files, mic, location, SMS, call logs, notifications
- apps that can draw over other apps
Step 4 - Check activity:
- battery use with the screen off
- data sent in the last 7 days, per app
- any app that uploads at night (1am-5am)
Step 5 - For every suspicious app, pull the APK, decompile it (jadx) and look for:
- server URLs and IP addresses
- code that reads the gallery, takes screenshots or logs keystrokes
- upload timers
Step 6 - Give me a report:
- a table: app / risk level (high, medium, low) / why
- for each high-risk app: what it can see, where it sends data, when it was installed
- a safe removal plan: what to save as evidence first, how to turn off device admin, how to remove it, which passwords to change and from which device
Explain everything in simple words. Don't remove anything until I say "remove"
i just built andrej karpathy's second brain in claude code... 1 folder, 3 agents, and my AI hasn't asked "who are you?" in 23 days
before this i lost 5 minutes every chat explaining who i am. 4 chats a day - 10 hours a month
karpathy's idea is simple: AI shouldn't meet you from zero every time. it should keep a wiki about you and update it itself
so i split it into 3 agents:
the collector saves every link, PDF and voice note i drop - raw, never edited
the writer reads each new source and rewrites every page it touches, 11 on average
the critic wakes up at 3:00 am and hunts for old facts, fights between pages and notes nothing links to
link → collector → writer → wiki → my question → answer saved as a new page → back to the wiki
the part nobody talks about: my questions feed it too. 184 of 486 pages came from answers, not from sources
23 days in: 214 sources, 486 pages, 1,904 links between them
the critic flagged 17 fights between pages. 4 of them were things i believed about my own money that were just wrong
every page starts with 3 lines: the claim, the source, the date it was last confirmed. 30 days with no new proof - it turns grey
opus 5.5 does the writing, haiku 5.5 collects and checks. $0.41 a day, $9.43 for all 23 days
the big mistake: dumping your whole life in on day 1. i did, and in 48 hours 70% of the wiki was noise
start with 1 topic you actually work on. setup took me 20 minutes: 1 empty folder, karpathy's page as the rules file, 3 agents in claude code
AI with no memory is a smart stranger. AI with a wiki is a partner that remembers ↓
Before you spend another $200 on Claude, read these 27 lines.
ANTHROPIC tested 30 CLAUDE.md files, and the shortest one won.
4 minutes to copy the setup.
Claude finished 9 out of 10 tasks without a second prompt.
The 600-line “ultimate setup” needed help on 8.
I double-checked the results.
Imagine paying for Claude every month, then making it worse with your own instructions.
Read the file and the article below.
Holy shit, Claude Opus 5.5 + Jev is a cheat code for your job.
One thinks like your smartest coworker. The other makes every small call in between, in 0.1 seconds.
Office workers lose 11 hours a week to email. Jev can clear 3,000 of them for 21 cents.
One guy fixed the links on all 586 pages of his site. 45 seconds, 21 cents.
Another found 700 leads and wrote the outreach in 40 seconds for 9 cents.
In Jev's own tests, 1,000 calls took GPT 2 hours and $13.88. Jev took 2 minutes and 8 cents.
Opus 5.5 dropped Tuesday, costs 40% less than the last Opus, and Claude killed the 5-hour limit.
Jev isn't even 2 weeks old and people have already built 3,400+ things with it.
Your boss has no idea this stack exists. Neither do your coworkers, yet.
Save the article below while it still makes you look like a genius.
Sam Bankman-Fried's $500 million bet on Anthropic is about to be worth ~$50 billion.
That's a 100x, and he won't see a cent of it.
He made that bet in April 2022, a 30-year-old in cargo shorts who slept on a beanbag at the office.
Claude wasn't even out yet. Anthropic had $0 in sales.
By November FTX was dead, and $8 billion of customer money was missing.
Prosecutors say the $500 million for Anthropic came out of that money too.
In 2024 the FTX bankruptcy team sold the shares for $1.3 billion to pay people back.
Back then 2.7x looked like a great exit.
WSJ says Anthropic plans to go public this November at around $2 trillion.
At that price, those shares would be worth about $50 billion, nearly 40x what they sold for.
He'll watch the IPO from a prison in LA, 3 years into a 25-year sentence.
In April his X account posted a chart of FTX's old bets and 3 words: "many such cases…"
The bet needed 4 years to pay off. He lasted 6 months.
If you've ever been liquidated 5 minutes before the move, you know the feeling.
Being right is worth $0 if you're not around when it pays.
I hired an employee who works 24/7 on a $0 salary. I still can't find the catch.
It's Elon's Grok Bot running Jev, an AI most people haven't heard of yet.
Grok Bot + Jev = your inbox, leads and DMs done while you sleep.
You pay ChatGPT $20 a month and it just sits there until you type.
The catch must be the price. 1,000 decisions cost 8 cents. Your coffee costs more.
Must be a scam then. Its creator spent 4 years inside OpenAI and just raised $40 million.
Or it's just hype. I ran my 3,412 leads through it. 15.7 seconds, 41 cents.
Office workers lose 11 hours a week to email. Fix that, then invoice their boss.
Every step is in the article below. You can copy it tonight.
Watch the video, then read it before the catch shows up.
A 4-cent AI just made your $20 ChatGPT look stupid.
Let me explain in 30 seconds.
ChatGPT is the ex who answers "yes or no?" with a 5-minute voice message.
It takes 8 seconds to say what fits in 1 word.
And you pay $20 a month to put up with that shit.
Jev is the one who texts back "yes" in 0.1 seconds. No drama.
It reads 7 novels for 4 cents. Writing costs $0, because it doesn't write.
In their tests, 1,000 answers cost $13.88 with GPT and 8 cents with Jev.
Now plug it into Elon's Grok Bot, which already works 24/7 on its own computer.
Grok does the clicking and typing. Jev sorts your emails, leads and DMs, 500 a minute.
Jev's creator says over 99% of AI will be used by software, not people.
It came out 8 days ago, and most people are still typing "please" to their ex.
One talks for 8 seconds. The other does it in 0.1.
Watch the video, then read the article below before this is normal in 2027.
Elon Musk's Grok just did a full day of sales work in 16 seconds. It cost 41 cents.
It went through 3,412 people on X, LinkedIn and YouTube and picked the 186 most likely to buy.
The rest were stuff like bots, students and lofi channels.
Then it wrote each of the 186 a message. All you do is click send.
Companies pay a sales rep $155,000 a year to do this by hand.
The trick is a new AI called Jev that plugs into Grok. It came out 7 days ago.
It was built by one of the guys who made ChatGPT, and almost nobody has access yet.
Setup takes 9 minutes.
Watch the video, then read the article below.
If you still look for clients by scrolling LinkedIn, this video is going to hurt.
A week-old AI called Jev just did 6 hours of that work in 15.7 seconds, for 41 cents.
The guy behind it helped build ChatGPT. And it can't write a single sentence.
Investors threw $40 million at it last week.
Plug it into Grok Bot and you get a sales assistant that never sleeps.
It checked 3,412 people. 95% were junk: bots, recruiters, students, lofi channels.
It found the 186 most likely to pay you and wrote each one a message.
Companies pay a sales rep $155,000 a year to do this by hand.
I keep rewatching it, and it still feels like cheating.
Setup takes 9 minutes. Read the article below before your competitors do.
You use Claude wrong.
The man who built Claude Code just showed how he actually uses it.
He gave this talk to Anthropic's team. Thank god my friend at Anthropic sent me the recording.
• 00:00 - The right way to use Claude
• 08:33 - How Anthropic prompts Claude
• 16:19 - Making $100k+/month with Claude
I watched it last night. Realized I've been using Claude like a toy.
Posting it below. It was never supposed to get out.
Andrej Karpathy joined Anthropic two months ago.
Yesterday my friend on his team sent me the Claude.md file he actually uses.
It completely changed how I work with Claude.
From the very first message, the difference was obvious.
With this file, Claude finally stops fighting me and starts working exactly the way I need it to.
Bookmark it before it gets taken down.
Read it now, then check the article below.
Delete every “How ChatGPT works” video you saved.
Stanford already gave away its $65,000 ChatGPT lecture for free.
• 25:30 - ChatGPT vs Claude vs Gemini
• 59:56 - Why your prompts suck
• 1:24:53 - How ChatGPT actually works
A Stanford professor explains ChatGPT properly. Not the Twitter version. The real one.
You use 10% of ChatGPT and Claude. This video gets you to 100%.
AI gurus will spend the next year selling this lecture back to you.
Watch this today, then use the guide below.
Andrej Karpathy put everything he knows about LLMs into this unreleased lecture.
"You don’t need better prompts – you need loops."
In 54 minutes, he teaches you the skill Anthropic pays $500K for.
You probably don’t have 54 minutes right now. Don’t lose this in the feed.
Watch it tonight, then use the guide below to build your first loop.
Stanford students pay $67,731 a year to learn this.
This is how they use ChatGPT when nobody’s watching.
A two-hour course from the best AI engineer alive.
Ten minutes in, I realized I was using it like everyone else.
02:51 — The prompt they use.
In a year, everyone will use ChatGPT like this.
Watch this tonight.
Stanford just leaked the FULL LLM course it only gives to premium students.
Bootcamps charge $20,000 and still teach less than this.
00:42 - How LLMs Work
06:15 - Text Into Tokens
39:00 - Transformer Internals
It's useful if you use ChatGPT every day.
More useful if you want a $750,000-a-year job at OpenAI.
My friend at Stanford sent me this mp4 and said not to post it.
Watch this tonight, then do the guide below.
Andrej Karpathy just dropped his first course since joining Anthropic.
90 free minutes to become an AI engineer before 2027:
19:13 - Getting Hired Without a Degree
30:31 - How Claude Actually Works
52:13 - Your First AI Agent
Bootcamps charge $20,000 to teach less than this.
You probably don't have 90 minutes right now. Don't lose this in the feed.
Watch it tonight. Then read the guide below and build your first loop.
Don't waste 2 years learning ChatGPT.
The man who created ChatGPT tells you what to learn instead.
69 minutes. Free.
17:04 - You don't need prompting anymore
34:08 - What to learn instead
51:12 - Build your first loop
Writing prompts is the old job. Building loops is the new one.
Watch it. Then build your own loop with the guide below.