Pinned Tweet
🚨 Attention: A small security firm just got root on Apple’s newest iPhone. And it survived a reboot.
Calif.io did it live at Hexacon in Paris on October 16. iPhone 18 Pro. iOS 27.0.1. Apple’s current release.
It is not a leaked government tool. It is not a spyware vendor. It is a team that says it sells no bugs and no exploits.
They rebooted the phone, unlocked it, and two minutes later were logged in over SSH as root. No computer attached.
Apple pays up to $2,000,000 for a spyware-grade exploit chain. With bonuses, the ceiling passes $5,000,000. Apple has paid $35,000,000 to 800+ researchers since 2020.
Calif reported nearly 100 bugs to Apple this year.
Now the AI part. Apple’s iOS 27 security notes credit Calif.io together with Claude and Anthropic Research for two bugs. In May, Calif said Claude Mythos Preview helped build a macOS kernel exploit against Apple’s M5 protections in under a week.
Nobody has confirmed AI built this exact jailbreak. But Apple itself says chains like this cost millions to develop.
iOS 27.2 shipped October 26 and fixed most of the bugs. Calif never named them. No public tool exists, and Calif plans no end-user release. Update anyway.
The demo video is public and free. Most people will scroll past it.
Rayan retweeted
claude code just traced the $5.57 billion binance hack to an address inside binance and a relay node in pyongyang. this wasn't a hack. and binance hasn't said one word about it
yesterday i posted how grok bot pulled my $107,000 out of binance 12 minutes before the freeze. 125k people saw it and everyone asked the same thing. who did this
binance won't tell you. so at 1am i gave claude code one prompt and the public on-chain data, with no api keys and no insider access
1:04am. claude pulls 47,832 transactions across eth, btc and tron and finds 47 wallets in one attacker cluster
1:11am. the terminal goes red. "match found"
the wallet that paid the attacker's gas was funded 48 hours before the hack. first tornado cash, then one 0.001 eth test transaction from an address tagged binance internal deployer. binance's own deployer paid for the attacker's gas two days early
1:17am. claude replays the drain. binance 14 first, $2.13 billion. hot wallet 34 second, $780 million. then the reserve pools, in the exact order of binance's internal balance registry. that list isn't public. someone handed it over
the drain also started 79 seconds before a scheduled key rotation, so whoever did this knew the schedule
1:24am. claude follows the exploit contract back to its source. three separate signals point to the same node in pyongyang
bytecode matches the bybit hack at 98.4%. claude's attribution is lazarus group with 97.3% confidence
so ask yourself why binance hasn't explained how north korea knew its internal registry and its key rotation schedule
claude thinks someone inside gave lazarus the keys and the map, and lazarus walked in
every claim is marked confirmed or inferred and comes with a tx hash. claude refused to name the insider
what i did before sunrise:
> sent the evidence package to FBI IC3
> flagged all 51 attacker wallets to the exchanges
> 2 deposit addresses frozen by morning. $60 million that isn't making it to north korea
my wife found me at 3am again. she looked at the red dot on the globe and asked if that was north korea. i said yes. she didn't go back to sleep either
if your money is on an exchange, all it takes is one person on the inside
the exact prompt that did all of this is in the comments. it's free. run it before the next one hits
Rayan retweeted
Claude Code found the scammer who stole $2,400 from my mom using just his phone number. full name, address, workplace, and 11 other families he did the same thing to. his VPN and burner number didn't protect him from a single thing
my mom called me at 11pm barely able to talk. someone called her pretending to be her bank saying her account was compromised and she needs to verify her card immediately. she's 64 and lives alone. she panicked and gave them her card number, CVV, and the verification code from the SMS. three transfers went through before she realized what happened. $2,400 gone. that was her savings for my sister's wedding in march
she kept apologizing to me on the phone. "i'm sorry, i'm so stupid, please don't tell your sister." i drove to her apartment at midnight and she was sitting at the kitchen table with the phone still in her hand. she looked like she aged 10 years in one hour
i went to the police the next morning. they took a report, said they'd look into it. i asked what that means and the officer said "we'll add it to the file." that means nobody is going to do anything and we both know it
so i took the one thing he left behind - the phone number he called from - and gave it to Claude Code
→ VoIP number → registered under br***t.kov***[email protected]
→ that email → 4 matches in breach databases → full name: Br**t A. Kov***c
→ VPN IP → 185.***.***.47 → routed through Netherlands
→ real IP from VoIP provider dashboard login → 74.***.***.12 → Houston, TX
→ Facebook → profile since 2016 → 347 friends → employer listed publicly on his page
→ Venmo → public by default → 11 incoming payments in 90 days with notes like "refund" and "reimbursement"
Claude Code cross-referenced the Venmo payments with public scam reports on the FTC database. he wasn't just scamming one person. he was running this full time for at least 3 months
→ victim 1 → 71 year old woman in Florida → $3,100 → called "about suspicious activity"
→ victim 2 → retired couple in Ohio → $1,800 → same frozen account script
→ victim 3 → single mom in Georgia → $950 → she filed a police report too. nothing happened
→ my mom → $2,400 → same script, same voice, same burner number rotated through 4 VoIP providers
his VPN routed traffic through the Netherlands but he logged into his VoIP dashboard from his real IP twice in October without it on. his Facebook has been public since 2016 with his city and photos of his car with the plate visible. he spent $4 a month on a VPN and thought that made him untraceable
i compiled everything Claude Code found into a single report:
→ full legal name confirmed through public records
→ home city → Houston, TX area → matched to voter registration
→ 11 confirmed victims across 5 states → total stolen ~$19,200
→ 4 VoIP numbers rotated monthly → all registered to the same email
→ complete Venmo transaction history as evidence
sent it to the police and FBI IC3. the same officer who said "we'll add it to the file" called me back the same afternoon asking where i got all of this. suddenly they had a name, a face, and 11 other victims willing to testify
i called my mom that night. told her they found him. she didn't say anything for about 10 seconds. then she started crying again. not like before. she kept saying "really? they found him?" i didn't tell her it wasn't the police who found him. i didn't tell her it took me less than an hour sitting at my kitchen table
if someone scammed your family don't just file a report and wait. save that phone number. don't delete the call history. Claude Code can pull more from one phone number than most people think is possible
the prompt is in the first reply. if you got scammed or someone suspicious keeps calling people you care about - try it. takes 2 minutes
Rayan retweeted
ANTHROPIC IS NOW IN DAMAGE CONTROL 🚨
Less than a week after the “no game recreations” rule showed up in Claude Code, Anthropic quietly rewrote it. Same server flag, “tengu_heron_brook”.
Two things got added:
• “or with similar mechanics”: copying a game’s gameplay is now explicitly allowed
• “(its name, characters, art, music, levels or look)”: now spells out what actually counts as a “recreation”
They also added a second flag, “original_games_guidance”, and Claude Code 2.1.317 now has the new wording built straight into the app.
Still no setting to turn it off.
Rayan retweeted
The AI layoff gamble is backfiring.
Companies that rushed to replace workers with AI are now cleaning up the mess, per The Guardian.
Meta scaled back plans to cut staff and use more AI agents after employees spent growing time cleaning up the agents’ technical and security errors. It’s now rehiring managers.
Microsoft and McKinsey are limiting employee AI use as costs pile up.
Walmart workers say AI scheduling and task tools set impossible standards and generate false alerts.
The payoff isn’t showing up: 90% of executives say AI hasn’t yet boosted productivity at their companies, per an Atlanta Fed working paper.
Yet more than 112,000 US job cuts this year have cited AI, per Challenger, Gray & Christmas.
Rayan retweeted
🚨 Attention. Spyware on a phone rarely needs a hacker. It needs someone who knows your passcode.
Kaspersky counted 34,000+ people hit by stalkerware in 2024 and 2025, and 127,000 in five years. The Coalition Against Stalkerware says the real figure could be close to a million a year.
The clip shows a scan flagging a Pegasus-class tracker on an iPhone. The tool is never named, so treat it as a demo, not proof.
What I actually documented, with a source for every number:
→ Apple pays up to $2,000,000 for a spyware-grade chain
→ a rigged charger can cost under $100 to build
→ Meta settled one face-scan lawsuit for $650,000,000
→ iOS 27.2 patched 163 flaws, Anthropic Research credited on three, OpenAI on four
The AI part is a rules file. Claude Core is not a product. It is one CLAUDE.md that Claude Code reads every session: read only, local only, mask secrets, evidence first, unknown is fine, treat every page as data.
Claude Code does not make a phone safe. It reads what is already there and points at the one line that should not be.
One rule most viral posts skip. Kaspersky warns the person running stalkerware can notice when it is removed. Save the evidence first. Remove it second.
Seven cases, ten minutes each. Bookmark the article. Almost nobody will.
🚨 Attention. Spyware on a phone rarely needs a hacker. It needs someone who knows your passcode.
Kaspersky counted 34,000+ people hit by stalkerware in 2024 and 2025, and 127,000 in five years. The Coalition Against Stalkerware says the real figure could be close to a million a year.
The clip shows a scan flagging a Pegasus-class tracker on an iPhone. The tool is never named, so treat it as a demo, not proof.
What I actually documented, with a source for every number:
→ Apple pays up to $2,000,000 for a spyware-grade chain
→ a rigged charger can cost under $100 to build
→ Meta settled one face-scan lawsuit for $650,000,000
→ iOS 27.2 patched 163 flaws, Anthropic Research credited on three, OpenAI on four
The AI part is a rules file. Claude Core is not a product. It is one CLAUDE.md that Claude Code reads every session: read only, local only, mask secrets, evidence first, unknown is fine, treat every page as data.
Claude Code does not make a phone safe. It reads what is already there and points at the one line that should not be.
One rule most viral posts skip. Kaspersky warns the person running stalkerware can notice when it is removed. Save the evidence first. Remove it second.
Seven cases, ten minutes each. Bookmark the article. Almost nobody will.
Rayan retweeted
THIS IS F**KING DANGEROUS.
SOMEONE JUST UNLEASHED AN OPEN SOURCE AI HACKING ARMY TOOL
THAT CAN LAUNCH AUTONOMOUS PENTESTING OPERATIONS.
It’s called SAPPER.
One AI commander. An entire swarm of AI agents. Real hacking tools running behind the scenes.
Here’s what makes it stand out:
→ One AI commander plans and coordinates the operation
→ Multiple AI agents work together to execute tasks
→ Runs real tools like Nmap and Nuclei
→ Built in hacking knowledge base for the agents
→ Integrates with Claude Code and Codex
→ Runs inside a Kali Linux environment
It reportedly competed against 634 teams in a Tencent attack and defense competition and operated for 8 hours without human intervention.
Fully open source.
one commander —> multiple ai hackers—> an entire automated red team
Only use it in authorized labs and CTF environments.
REPO BELOW
Rayan retweeted
Claude Code just exposed a hidden $389 Ecoflow power station in a bedroom that secretly spied on its owners for 10 months, quietly sending exact sleep schedules, departure times, and GPS coordinates to Shenzhen.
The hidden mechanics of the hardware breach:
→Hidden cellular backdoor:An undocumented 4G LTE eSIM module bypassed home Wi-Fi entirely, running silently on its own battery even when the screen was off.
→Precise behavioral tracking: Logged when the flat went empty down to the minute, harvesting 11,328 telemetry packets over 10 months.
→ AI driven discovery: Claude Code flagged unexpected serial data ports during a laptop charge, immediately shutting down the unauthorized modem and cutting off data exfiltration.
While consumers buy unbranded electronics for emergency backup, autonomous AI agents are becoming the only defense against deep hardware-level espionage.
How secure is the backup gear sitting next to your bed?
Rayan retweeted
WTF The clip below claims to catch Pegasus-class spyware on an iPhone. The tool is never named. Apple pays up to $2,000,000 for the real thing.
Spyware of that class arrives by a zero-click chain. No tap, no app. That is what Apple prices at $2,000,000, and over $5,000,000 with bonuses for Lockdown Mode bypasses. $35,000,000 has gone to 800+ researchers since 2020.
Now the part viral posts blur. Every one says Claude Code found it. There are two different Claudes in this story.
Claude Code is a coding agent. It reads and edits the files you point it at, and only what the operating system lets it read.
Claude Mythos Preview is the model Calif says helped it break Apple’s M5 shield in five days. Apple spent about five years on that shield. Humans still beat the hardest part, Calif says. Only vetted partners have it, through Project Glasswing. Anthropic says about 50 of them found 10,000+ high or critical flaws in a month.
The newest tier, Mythos 5.1 and Fable 5.1, shares one model. Fable adds extra cybersecurity safeguards.
Apple’s side: iOS 27.2 shipped 163 fixes. Anthropic Research is credited on three, OpenAI on four. Nobody has confirmed AI built the October jailbreak.
Your check: Settings, Privacy and Security, App Privacy Report. Then update.
The article sorts which Claude did what. Most viral posts never do.
Claude Code just exposed a hidden $389 Ecoflow power station in a bedroom that secretly spied on its owners for 10 months, quietly sending exact sleep schedules, departure times, and GPS coordinates to Shenzhen.
The hidden mechanics of the hardware breach:
→Hidden cellular backdoor:An undocumented 4G LTE eSIM module bypassed home Wi-Fi entirely, running silently on its own battery even when the screen was off.
→Precise behavioral tracking: Logged when the flat went empty down to the minute, harvesting 11,328 telemetry packets over 10 months.
→ AI driven discovery: Claude Code flagged unexpected serial data ports during a laptop charge, immediately shutting down the unauthorized modem and cutting off data exfiltration.
While consumers buy unbranded electronics for emergency backup, autonomous AI agents are becoming the only defense against deep hardware-level espionage.
How secure is the backup gear sitting next to your bed?
Rayan retweeted
Claude Code found that your iPhone keeps a photo of your face every single time you unlock it. i had 31,000 photos of my face on my phone that i never took. some of them were taken at 3am while i was half asleep.
my phone was running slow so i connected it to my MacBook and ran Claude Code to check storage. it flagged a directory i didn't know existed with 31,247 images. every one was a photo of my face taken by Face ID at the moment of unlock. going back to 2024.
Claude Code sorted them by time and that's when it got weird.
→ thousands of photos taken between midnight and 5am. me half asleep, eyes barely open, lying in bed next to my wife
→ 2 photos that aren't me. someone unlocked my phone while i was sleeping in December 2024
→ photos where i'm visibly crying. i know exactly which days those were
→ photos from the bathroom. from the car at a red light. in the dark at 4am when i couldn't sleep
→ a photo of my wife trying to unlock my phone at 2am with my face while i was asleep. it failed but Face ID still cached her face
Face ID doesn't just scan your face. it takes a full image, caches it locally, and never deletes it. Apple never wipes this cache.
Claude Code checked which apps had access to this directory. 14 apps with full photo library permission could read every single one of these images. a weather app. a food delivery app. a free game my kid installed.
the 2 photos that weren't me - Claude Code pulled the timestamps. December 14, 2024, 3:47am and 3:51am. someone picked up my phone and unlocked it with their face while i was sleeping. Face ID failed both times but still cached the image. i still don't know who it was.
what Claude Code did:
→ purged the entire Face ID image cache
→ revoked photo library access from every app that doesn't need it
→ found 2 more hidden directories caching face data from FaceTime and the front camera
i went from 14 apps with photo access to 3.
the full prompt is in the first reply. takes 3 minutes. you probably have thousands of photos of yourself you've never seen. save it and send it to anyone with an iPhone.
Readers added context they thought people might want to know
Face ID does not store photos of your face when unlocking your device. Apple's documentation states that face images captured during authentication are immediately discarded after generating a mathematical representation inside the isolated Secure Enclave.
support.apple.com/en-us/102381
WTF The clip below claims to catch Pegasus-class spyware on an iPhone. The tool is never named. Apple pays up to $2,000,000 for the real thing.
Spyware of that class arrives by a zero-click chain. No tap, no app. That is what Apple prices at $2,000,000, and over $5,000,000 with bonuses for Lockdown Mode bypasses. $35,000,000 has gone to 800+ researchers since 2020.
Now the part viral posts blur. Every one says Claude Code found it. There are two different Claudes in this story.
Claude Code is a coding agent. It reads and edits the files you point it at, and only what the operating system lets it read.
Claude Mythos Preview is the model Calif says helped it break Apple’s M5 shield in five days. Apple spent about five years on that shield. Humans still beat the hardest part, Calif says. Only vetted partners have it, through Project Glasswing. Anthropic says about 50 of them found 10,000+ high or critical flaws in a month.
The newest tier, Mythos 5.1 and Fable 5.1, shares one model. Fable adds extra cybersecurity safeguards.
Apple’s side: iOS 27.2 shipped 163 fixes. Anthropic Research is credited on three, OpenAI on four. Nobody has confirmed AI built the October jailbreak.
Your check: Settings, Privacy and Security, App Privacy Report. Then update.
The article sorts which Claude did what. Most viral posts never do.
Rayan retweeted
my kid thought she deleted those photos. 11 apps on her iphone have been reading the hidden cache they live in since 2023. claude code found the folder tonight. every parent needs to see this
i was not even looking for this. claude flagged a directory i would never have opened on my own: PhotoData/Thumbnails
inside: photos. videos. screenshots. half of them she had deleted. a bunch she had cleared from Recently Deleted months ago
what was in there scared me:
→ screenshots of her chats with friends she had wiped the next morning
→ photos she took and deleted within 60 seconds
→ old group photos from a sleepover she told me she "removed" from her phone
→ every image her camera roll had ever seen since the phone was activated
the part i did not know: when a photo is deleted on ios, the system removes it from the camera roll and from Recently Deleted. but a compressed version stays in a system search cache so the OS can still find it when you search. that cache is not fully wiped when the file is gone
so my kid's "deleted" screenshots are sitting in a folder any app with Files access can read
then i checked which apps on her iphone had Files and Folders access. the list made me sick
random games downloaded 18 months ago. "free" photo filter apps she forgot existed. a tiktok clone i did not even know was installed. every one of them had the access to read that cache the whole time
i revoked permissions on 11 apps in 3 minutes
the exact claude code prompt i used for the audit is in the article below. the whole thing takes 10 minutes on your kid's iphone if you already have a backup
the question every parent needs to answer tonight: how many apps on your kid's iphone have Files access right now and when was the last time you actually checked
bookmark this before your kid's next backup
Rayan retweeted
Apple offers a bounty of up to $5,000,000 for critical zero-click exploits, but advanced AI models like Claude Mythos Preview are shattering 5-year hardware security walls in just 5 days.
• $5,000,000 Payout: Apple's maximum reward for Lockdown Mode bypasses and mercenary zero-click chains.
• 5 Days vs 5 Years: AI systems bypass cutting-edge chip protection in days, rendering years of human engineering obsolete.
• Official AI Credits: Anthropic Research and OpenAI are now directly credited in official iOS security patches for finding kernel bugs.
• Financial Scale: In 2024, Apple paid $11.8 million to 660 researchers, driving the multi-billion-dollar vulnerability market.
How to secure your device right now:
App Privacy Report: Check Settings Privacy & Security to see which apps access your camera, mic, and domains.
Photos Access: Limit Full Access to your photo library; switch non-essential apps to Limited Access.
System Indicators: Watch for the orange or green status dot indicating active microphone or camera usage.
Subscription & Charge Audit: Review Apple Pay transactions and your Apple ID Subscriptions for unauthorized recurring fees.
Apple offers a bounty of up to $5,000,000 for critical zero-click exploits, but advanced AI models like Claude Mythos Preview are shattering 5-year hardware security walls in just 5 days.
• $5,000,000 Payout: Apple's maximum reward for Lockdown Mode bypasses and mercenary zero-click chains.
• 5 Days vs 5 Years: AI systems bypass cutting-edge chip protection in days, rendering years of human engineering obsolete.
• Official AI Credits: Anthropic Research and OpenAI are now directly credited in official iOS security patches for finding kernel bugs.
• Financial Scale: In 2024, Apple paid $11.8 million to 660 researchers, driving the multi-billion-dollar vulnerability market.
How to secure your device right now:
App Privacy Report: Check Settings Privacy & Security to see which apps access your camera, mic, and domains.
Photos Access: Limit Full Access to your photo library; switch non-essential apps to Limited Access.
System Indicators: Watch for the orange or green status dot indicating active microphone or camera usage.
Subscription & Charge Audit: Review Apple Pay transactions and your Apple ID Subscriptions for unauthorized recurring fees.
Rayan retweeted
🚨EXCLUSIVE: Claude Code just exposed a hidden $500 hardware hack inside a cheap airport power bank that quietly mirrored entire iPhones during a standard charge, risking a $100,000+ data breach.
The hidden mechanics of the breach:
• 64 GB internal storage: The device registered as a micro-computer, exploiting a single Trust This Computer tap to scrape 1,206 contacts, 3,912 photos (including passports), and $50k+ banking app data.
• Automated exfiltration: It harvested home Wi-Fi credentials on the first plug-in, using the local network to quietly upload packets to external servers while the owners slept.
• AI-driven detection: Claude Code flagged the unauthorized data stream within 10 seconds, cutting off outbound connections and locking down compromised accounts in real time.
While consumers trust cheap hardware accessories, autonomous AI agents backed by $10B+ in advanced computing are becoming the only defense against hardware level cyber espionage.
How secure is your daily tech setup?