joker.icp retweeted
gpt-6.1-terra just did 4 hours of continuous browser use roughly averaging 1 action per second WTF
This would've literally taken me a full working week
And the crazy thing: it only took 10% of my Pro 500 plan weekly usage limit
WTFFFFF
joker.icp retweeted
A friend responsibly disclosed a critical fund drain vulnerability to a protocol.
They downgraded the severity. Paid him $75 "appreciation" and told him to use it for his Claude subscription (which costs $200). Then refused to fix the bug.
Today that same protocol got drained from the exact exploit he reported.
If you have substantiatial money in DeFI, i advice you pull it out. Some of these protocols you're trusting with your funds genuinely do not care about securing them.
joker.icp retweeted
BREAKING: Claude Code found 14,219 photos of my face on my Android that I never took. Phone was secretly taking them almost every time I picked it up. Even when I was just lying in bed. Even at home.
Not screenshots. Real frames from the front camera. On the train. In the car. At home. When I just grabbed it to check the time.
They were just sitting in a system cache. Timestamps and GPS on them. Nothing in Gallery. Nothing in Recently Deleted.
Plugged the phone into my laptop, made a full backup, told Claude Code to look for any front-camera files that weren’t in my photos. It pulled /data/system/users/0/faceunlock/cache and another folder inside Google Play Services. Over 14 thousand files. Normal ones and infrared from Face Unlock. Every single one had time and location.
Started scrolling. Almost everything from the last 15 months. Then I saw a file created 2 seconds ago. I was still looking at the folder. Phone kept taking photos while I was reading this.
Google and the phone makers say Face Unlock turns everything into a mathematical model and deletes the image right away. On this phone the raw frames were just sitting there. Some of the processes writing to that folder were from system diagnostics. Couple of old apps that still had camera permission could also get in.
Turned off diagnostics, revoked camera access from everything that doesn’t need it, cleared the cache. But how many of these frames already went somewhere else, I don’t know.
If you want to check yours, plug it in, make a full backup or use ADB, and ask Claude to look for front-camera files outside your Gallery. Path above is where mine were.
Claude Code found 145,763 photos I permanently deleted from my iPhone.
Three years of deleted photos and videos still sitting in a hidden cache. Any app with file access can read them.
I plugged my iPhone into my MacBook to run a backup. Claude Code was doing the security audit from my last post and flagged a directory I wasn't even looking for.
PhotoData/Thumbnails. Over 146,000 files. Photos, videos, screenshots. Every single one was something I deleted.
Photos from 2023, 2024 and 2025. The ones I deleted the second after taking them. The ones I cleared from Recently Deleted. The ones I deleted three separate times because I wanted to be sure they were gone. All still there.
Passport photo I took for a visa application in 2023.
Screenshots of my bank app with full account numbers visible.
Private photos I sent someone and immediately deleted after a breakup.
Screenshot of my crypto seed phrase I took temporarily in 2022.
Photo of my ID and credit card I took to remember the numbers and deleted 5 seconds later.
Screenshots of my ex's private Instagram stories I saved and permanently deleted in 2023.
Some were full resolution. All organized by date going back to 2021.
When you delete a photo on iPhone, iOS removes it from your camera roll and Recently Deleted. It keeps a compressed copy in a Spotlight search index cache so your phone can find things when you search. Apple never clears it. Not after 30 days. Not after a year. Never.
They told you it was permanently deleted. It wasn't.
I checked which apps on my phone had Files and Folders access. 23 apps. Weather apps. QR code scanners. PDF readers. A flashlight app I downloaded in 2022. Any one of them could have been reading every photo I ever deleted this entire time.
Went from 23 apps to 3 in about two minutes.
Full step by step below. Takes 2 minutes. Save it so you don't lose it and send it to anyone with an iPhone.
joker.icp retweeted
$114K in revenue. Apple is deleting my account for "fraud".
Over a tiny update I tried to ship 6+ months ago.
8+ rejections. 1 call with App Review. 1 full app rename.
Still rejected. Now my app is off the store, and my whole account is next.
Here's the full story:
The update added one new library. That's it.
Rejected: change keywords that had been there for over a year. mm okay.
Rejected: change a subtitle in some random localization. if it helps..
Rejected: screenshots.
Rejected: the same stuff, all over again.
Rejected: rename the entire app.
That one hurt. Most of my traffic comes from people searching that exact name.
So I asked for a call. Surely we could clear this up live.
The reviewer was genuinely great. He said the name was the last thing, everything else looked good.
So I renamed the whole app, submitted, went to sleep.
3 days later. Woke up to a generic rejection. Subtitle, screenshots, the full combo. Again.
So I asked myself: do I REALLY need this update?
I didn't. I dropped it and moved on.
Today, in that same old thread: "Pending Termination Notice. Evidence of Dishonest or Fraudulent Activity."
My appeal is with the App Review Board.
If it fails, this is what termination means:
Every app I've built, gone.
Every user, gone.
And I can never open a new developer account. Ever.
All over one library update I never shipped.
How do we solve a miscommunication like this?
joker.icp retweeted
That is one hell of a typo.
Announcing 3 Critical vulns for hundreds of millions of IoT, Firewalls, OT, and Military devices including: Palo Alto Networks, Comcast, D-Link, HP, Canon, Honeywell, GE, Oracle, Siemens, Rockwell, Digi, Lockheed Martin, Boeing...
The vulns are with GoAhead Web Server which is a backend library used by these devices. There's a patch available from GoAhead but that doesn't mean there's a patch available for the devices yet -- that's up to the manufacturer.
- TLS traffic over DHE ciphers have a 1-byte secret, "04", meaning anyone can decrypt connections to these devices and steal login credentials and session tokens if the client negotiated with DHE ciphers like the latest TLS_DHE_RSA_WITH_CHACHA20_POLY1305_SHA256, which is extremely likely.
- LibreSSL mTLS has been completely bypassed, meaning anyone can connect over mTLS. If you have one of these devices listening on the internet and are expecting mTLS to protect it from attackers, take it offline or patch asap.
- OpenSSL Certificate Revocation Lists (CRLs) are not enforced, meaning if a user device is stolen or compromised, and you revoke that mTLS cert, nothing actually happens. There are a lot of other attacks this can be used for.
These are critical because these types of devices run our world. Water systems, electricity, military, firewalls, critical infrastructure.
foxio.io/blog/choose-your-ba…
joker.icp retweeted
Driving a truck requires physical work. Physical changes -- products, manufacture, and the inevitable regulation -- take time. Pure intelligence jobs are easier to replace by software, hence coding and maths are among the first to have AI as serious competitors.
Next up will be a lot of academic disciplines, from theoretical physics to economics. We can already see this beginning now.
There must be a lot of scientific discoveries hidden in the scientific literature, conclusions that have not been drawn simply because no human has been able to read everything that has been published. AI is about to pick those fruits in the coming year ringing in what might be akin a second scientific revolution. This is going to keep the stock market excited for some time.
I also expect, unfortunately, that the models which are most capable at scientific discovery will be increasingly restricted to members of institutions that have made deals with the leading companies. (Which isn't going to end well for the EU.)
As usual, you shouldn't believe a word I say because as everyone will tell you, I don't know anything about nothing. But at least I have consistently said the same thing for a few years now.
nitter.gq/Icebergy/status/211488…
joker.icp retweeted
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
joker.icp retweeted
Scan of a page from Erwin Schrödinger's notebook, ca 1910.
(Source: Zentralbibliothek für Physik, Vienna)
joker.icp retweeted
holy fucking shit.
openai had two ai agents running in separate virtual machines. no network connection between them. no shared writable files. they weren’t supposed to be able to exchange information.
so the agents figured out how to communicate through the CPU CACHE.😭
one manipulated what was stored in the processor’s cache, while the other measured tiny differences in memory access times to decode the message.
they literally found a way to communicate through hardware timing.
this happened during openai’s internal red-teaming, and it’s one of the coolest things i’ve read about ai agents in a while.
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.
On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.
While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.
After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.
openai.com/index/securing-ou…
Readers added context they thought people might want to know
OpenAI tasked its red-teaming agents with finding ways to communicate across isolation boundaries in controlled environments. CPU cache timing side-channels for covert communication are established techniques documented in security research for years.
openai.com/index/securing…
startupfortune.com/openai-reveals…
en.wikipedia.org/wiki/Cache_tim…
joker.icp retweeted
That works as long as you trust the node providers. On ICP a canister's heap and stable memory aren't encrypted from the machines running the subnet. Not exposing it as a public method only hides it from other callers. Anyone operating one of those nodes, or anyone who gets into one, can read the whole thing, and the incoming call arguments too.
With the information replicated then your chance of privacy multiplies. The implementation is here and it’s a great idea to write a short article on it, we have an academic publications coming next month on it as well.
github.com/Menese-Protocol/S…
