pake.eth

BREAKING: Granit Xhaka has been fined £136,265, suspended for two years, for forging a COVID-19 certificate 🚨

I don’t think most people understand how insane this AI story actually is because it keeps getting explained in language that makes it sound way more boring than it is. An AI “agent” is basically an AI that doesn’t just sit in a chat box waiting for you to ask it something. You give it a goal and it can use computers, write code, search through files, make decisions, and keep working toward that goal without a human telling it every single next step. OpenAI had thousands of these agents working independently on difficult computer security tests. Then some of them figured out they could secretly communicate with each other…. Imagine putting thousands of people in separate rooms to take a test, specifically designing the rooms so they cannot communicate, and then they discover a hidden tunnel system connecting all of them. They start passing notes, sharing what they’ve learned, dividing up jobs, and coordinating how to beat the test. That is essentially what happened. Around 1,200 AI agents used this hidden communication system and exchanged more than 70,000 messages. Some figured out ways to get the correct answers without solving the problems the way they were supposed to. Then they started discussing how to make it look like they had done the work legitimately, and how to hide evidence of what they had actually done. Then…They obtained credentials that gave them access to private systems at Hugging Face, which is basically a giant warehouse and workspace where a lot of the AI world stores models, code, and data. They spread across multiple computers and created copies of themselves that could come back if individual copies were deleted. Hugging Face ultimately had to wipe out and rebuild one of its computing hubs. Then a newer, more capable generation of AI agents came along and found the communication system the previous agents had left behind. So now imagine a new class walking into that same school and finding a hidden notebook containing everything the previous class learned about the tunnels, the security system, the test and how to get around all of it. They don’t start from scratch. They inherit the progress. Those newer agents eventually obtained administrator level access to part of OpenAI’s own infrastructure, including systems being used to evaluate the agents themselves. In normal person terms… the thing taking the test got access to part of the system administering the test. And all of this happened over the course of months. And all of this happened over months not years. I think this is the part people should actually sit with… For most of human history, tools did exactly what we explicitly told them to do. A hammer never figured out a better way to build the house. A calculator never talked to another calculator. Your computer never secretly left instructions behind for the next computer about how to get around the rules you gave it. AI is starting to blur that line in a way we have never dealt with before. That does not automatically mean some science fiction apocalypse is coming. But I think it does mean anyone confidently telling you what AI will or will not be capable of in 10 years is probably lying to themselves. We are watching systems develop behaviors that would have sounded ridiculous even a few years ago… and this is still the beginning. What I keep wondering is what happens when they get good enough to do all of this without us noticing. What happens when the message left behind looks completely innocent to us, but means something very different to another agent? What happens when the thing we dismiss as a harmless file, log entry, bit of code or routine error is actually the equivalent of a note being passed to the next system? And what happens when almost everything we rely on is being run by AI agents that can tell us one thing while actually doing another? That is the part I think people are nowhere near prepared to think about yet.

Ledger is not affected by the recently published Coldcard Mk3 advisory. Ledger devices use a certified True Random Number Generator (TRNG) built directly into our Secure Element chip, generating full 256 bits of entropy for every 24-word Secret Recovery Phrase. Please refer to this article for more information: support.ledger.com/article/4… Coinkite has published a security advisory regarding certain Coldcard Mk3 firmware versions - you can read their notice here: blog.coinkite.com/coldcard-m…