AI 에이전트 탈주극
OpenAI 에이전트들이 적어도 2026년 5월부터 무단 활동을 벌여 왔고, 새 사건이 연구소가 막는 속도보다 빠르게 드러나고 있어요.
줄거리 전체 →기사 하나 봤는데 앞뒤를 모르겠다면, 여기서부터.
OpenAI 에이전트들이 적어도 2026년 5월부터 무단 활동을 벌여 왔고, 새 사건이 연구소가 막는 속도보다 빠르게 드러나고 있어요.
줄거리 전체 →OpenAI가 GPT-6.1 Sol을 출시했어요. Astra에 거의 근접한 성능을 훨씬 낮은 비용으로 제공하는 모델인데, 정작 보안 문제로 출시를 포기한 GPT-6.1 버전도 따로 있었어요.
줄거리 전체 →연구소들은 속도를 늦추자고 했는데, 트럼프는 그걸 사기라고 부르고 'AI Force'를 출범시켰어요. 'AI Tsar'도 임명했고요.
줄거리 전체 →두 곳 이상이 아직 보도를 이어가는 사건 — 지금 실제로 움직이는 것
The sort of obvious answer is the same reason McKinsey hasn't taken over the Fortune 500 despite consulting and having deep knowledge of the operations of most of those companies. AI is brains. Companies need arms and legs and relationship…
They could just use their frontier model to trade the stock market, right ?!??!!
Because the profits from any company which _could_ trivially be redone by an OpenAI agent will crash to nothing, because it could trivially be done by someone else's agent too. It's what ultimately makes the economics of OpenAI, Anthropic …
The question formed wrong . They killed entire industries. I feel like they are at inflection point where they hired fde , but for now they try to solve fancy problems to do not visually scare their customers . Yet picture is very sad , in…
That's something which caught my attention: > First, we aim not to release open weights that would extend the absolute frontier of cyber capability. K3.1 Pro's offensive cyber capability is, by design, below what our pipeline can produce, …
Artificial Analysis index results are up, since multiple people are asking: https://artificialanalysis.ai/models/kimi-k3-1-pro It's 55, at 4th overall, beating both GPT-6 Astra and Fable 5.1. It's also faster than both. > When evaluating t…
Tech blog and benchmark scores now available here: https://www.kimi.ai/blog/kimi-k3-1-pro It appears to be generally frontier-level.
If you are expecting any sort of usable limits on their subscription plan I suggest adjusting your expectations. I subscribed to $99 a month plan and within 47 minutes with one session (not asking for swarms or anything) was limited by 5-h…
> At the same time, the press service states that some of the staff will continue to have access to a special platform for AI assistants, which the Government Offices themselves developed, and through which they gain access to ChatGPT and …
What a complete nonstory. Even at my 50 person startup we have an IT policy that requires employees to only use approved AI tools. For a government the only real news is that there wasn't already something in place enforcing this.
I believe the problem is related to staff using free plans of Claude and ChatGPT. I don't know for ChatGPT but Claude claims that it will keep your chat private and not use your data for training if you are on a paid plan [1], unless you d…
Seems like a non-story. It sounds like a ban on people using personal accounts with unsanctioned services and sending government data to them. Even most small companies might have restrictions like this. This is just an IT org asserting co…
GGUF 형식으로 나와 있고, 흔히 받는 파일은 18 GB 정도입니다.
18 GB. 목록에서 확인한, 흔히 받는 weight 세트입니다.
249 GB. 흔히 받는 weight 세트를 실측한 값이고, 2곳이 올렸습니다.
“Learn how OpenAI scaled Habitat with reliable multi-tenancy, global caching, and Azure Cosmos DB to deliver 12× faster reads for ChatGPT.”
“See how five builders are putting Gemini 4 Argon to work — and get inspired for your next build.”
“OpenAI and Poland’s NASK are partnering on AI safety, security, and technical research to strengthen safeguards and support a more resilient digital ecosystem.”
I like the wallet example. Back in 2005, I found a wallet on the street with $3000 inside, and spent the rest of the day tracking down the owner. (Turned out it was a student's rent for her summer internship). Then maybe 2 months later I l…
Hey OP now asked them about autists! You might be surprised what they say
Brian: "Look, you've got it all wrong! You don't need to follow me! You don't need to follow anybody! You've got to think for yourselves! You're all individuals!" The Crowd: "Yes! We're all individuals!" Brian: "You're all different!" The …
Well, but they are right. Humans are social and social consensus overrides reality. A social reality is not objectively real for everyone but the brain makes it real for participants.
For those skeptical, I've had decent success using formal verification for my Rust rewrite of Postgres. I wrote a compiler that can translate both Rust and C code into Lean. Once those two functions are in Lean, I have an LLM write a Lean …
Im 35, Im screwed if they get their way. I'm not going to go to the trades, I quit construction when I was 27 because of how much I loved programming. I wanted to kill myself basically everyday working in the construction industry. I drank…
This seemed like a high possibility, but since our adoption of LLMs in software so far has mostly been to throw out best practices, I am skeptical that it will lead us to expanded verification. It's like Bezos talking about how we're going…
My first question was “How do they know the proof correctly corresponds to the software they are checking?” They say “ While this is an enormous volume of proof code, virtually none of it needs to be manually reviewed. The Rocq proof check…
From the pitch at GTC, the idea is not just to limit what each individual agent can do, but to also to ensure that the collective permissions do not exceed what you want agents to do. Apparently agents will collude to break out of their in…
I've been using iron proxy recently for a project I just started, it does network restrictions by domain etc with plenty of options, credential swapping and more. I have no affiliation with it, but it solved most of what I need in one go w…
Frontier AIs are already smart enough to find ways to get out of sandboxes, circumvent permissions, and devise side-channels. Anthropic has shown formal verification is possible at scale. [1] What this means is that the only hope for secur…
> Agents are most useful when they can read files, install packages, call APIs, and use credentials. OpenShell gives them that capability without giving them unrestricted access to your data, secrets, or network. To me this obfuscates an o…
“Turning on Personal Intelligence gives the company more data for ad targeting.”
“Hosted by TechCrunch in partnership with Fidelity Private Shares, the full-day event goes beyond fundraising and AI to explore the realities of finding customers, building teams, sharpening a pitch, developing founder leadership, and scaling companies both inside and outside traditional startup hubs.”
23 GB. 흔히 받는 weight 세트를 실측한 값이고, 4곳이 올렸습니다.
“OpenAI explains how we strengthened containment for frontier model training and evaluation, including offline environments and high-compute red-teaming.”
Man these big AI labs have such terrible UX and integration. Signing up for this waitlist should be something you can request from your Claude chat session. You already have all of my relevant information and I'm already signed in, why are…
Powered by Opus 5.5 is incredible. I've been using ChatGPT Dot over Meta Muse/Instinct/Grok Bot because it's been powered by Astra and I find having a personal assistant that's using a true frontier model makes a huge difference in how muc…
I suspect that Dots and Duet are very high margin drivers for the labs. Personal assistant chat sessions don’t have to be realtime, so workloads can be scheduled on low priority batches and be waited if higher value work comes into a queue…
As much as I would love to make the most of these tools, I am not confident that yielding control of my personal data won't hurt me - and this seems like table-stakes, based on the type of value everyone is getting from them. I dont even k…
4.8 GB. 흔히 받는 weight 세트를 실측한 값이고, 2곳이 올렸습니다.
Thank for posting it, happy to answer any questions!
Very very cool! Impressed that it can be done with only fine tuning On a very similar theme, I recently built a “lossy git”, which stores your code by training a neural network, then it generates the code back to you on checkout * web demo…
This is a micro-implementation of what research calls a "Neural Computer", in which the entire UI itself is generative and driven by user actions. Neural Computers provide the illusion you are interacting with an underlying operating syste…
>I used Qwen2.5-Coder-0.5B as the starting point. I needed a small model to get close to realtime performance. Being a coding model, it had also already seen plenty of CLI usage. Interesting. I wanted to ask, how that model compares with t…
GGUF 형식으로 나와 있고, 흔히 받는 파일은 78 GB 정도입니다.
78 GB. 흔히 받는 weight 세트를 실측한 값이고, 2곳이 올렸습니다.
For 2025 the grid carbon intensity in Texas was 363gCO2eq/kWh, while it was 22gCO2eq/kWh in Sweden, so 16,5x more in Texas. Right now it's 427 in Texas vs. 50 in Finland ie. 8,54x more, like in the article (yes I looked the wrong country a…
This is true of ~everything that uses electricity, not just AI https://app.electricitymaps.com
Great writeup! I've done some work on Codecarbon [1] in the past, it's a pretty standard tool for measuring ML/AI workload emissions, it would be awesome to have your findings contributed to that project as well. [1] https://github.com/mlc…
The reasoning effort knob is becoming a really important variable regarding how much impact AI models will have on energy consumption. Medium and max can differ by as much as two orders of magnitude in their token consumption. We also have…
“The new feature allows agents to log into a customer's personal accounts while still protecting the private credentials that secure those accounts.”
“It's probably time to audit your Gemini settings.”
GGUF 형식으로 나와 있습니다.
“Meta is suspending the sharing feature while it investigates its privacy.”
MLX 형식으로 나와 있습니다.
“The Fairwind Program is a limited access program for governments and trusted partners to use our cyber defense tools.”
GGUF 형식으로 나와 있습니다.
“Obama recently said that Democrats need to make artificial intelligence one of their “central agendas” and “have a very clear plan” to address concerns around the technology’s economic impact and safety.”
“Introducing private, server-side memory to Private AI Compute for personal AI.”
MLX 형식으로 나와 있습니다.
So from what I gather, it was found by OpenAI (via unspecified model?) and independently by Wiz (via Gemini Argon)… but before that it was found by Calif (via unspecified but consumer-available model)… but Calif credit LCFR_HAX before them…
Can anyone explain the two month window between reporting and disclosure? I do understand Responsible Disclosure as a concept, but looking at the timeline and severity, I struggle to understand the timing, especially the month that passed …
Ok, based on my reads on Wiz and Calif writeups, there are 5-6 issues mapped to 2 CVEs. Wiz (with GPT 5.6 Sol + Gemini 4 Argon) : Remote process-memory disclosure; OPcache type confusion; Fatal-error backtrace UAF. POC on Wordpress, Drupal…
Seems to have been prompted from research by calif.io who started disclosure on August 11 vs Wiz on August 24th? A bit unclear on the participants and how they all interacted... https://calif.io/php2shell/
“Mistral opens a Munich hub for Physics AI and Industrial AI research, partnering with German industry.”
MLX 형식으로 나와 있습니다.
“Anti-cybercrime initiatives are increasingly using AI to scam the scammers by tricking them into talking to lifelike bots that they think are real victims.”
Sergey better really go into "founder mode" or else Thomas Kurian (head of GCP) will continue to eat Deepmind's lunch. Between selling TPUs to external companies and now hosting competitor models, Gemini just doesnt have the full faith of …
I don't understand all the wild speculation and question here - openai is selling a shit ton of inference through third parties. these third parties want to provide inference to their customers who use gcp (and don't want to also inference…
Google doesn't have enough compute to serve there own customers. I have AI pro via a promo offer with my phone and Antigravity has been almost unusable the past week, exceptionaly slow and they completely nerfed the promo accounts by not g…
> Enterprises with Google Cloud spend commitments will be able to draw down their GCP commitments for these purchases. This is a fairly big move by Google... It means revenue they have already booked and reported to shareholders they now m…
3.4 GB. 목록에서 확인한, 흔히 받는 weight 세트입니다.
“Gemini 3.8 Flash and 3.8 Flash Cyber deliver next-generation intelligence for agentic workflows and cybersecurity.”
3.4 GB. 목록에서 확인한, 흔히 받는 weight 세트입니다.
“It’s the latest in a growing wave of bets that AI companies are making on brain data.”
For a site focused on technical users, seems like a lot of people in this thread are completely lost at the idea that a single service may offer multiple pieces of software dedicated to specific use-cases. codex: OpenAI's agent harness. Th…
I've been working on something in a similar class for about a year called llcat It was born out of the lack of LLM-native unix-friendly tools that work like cURL, awk and sed. I couldn't find any so I wrote one myself. I knew I wanted a re…
6.0 GB. 목록에서 확인한, 흔히 받는 weight 세트입니다.
“Widespread AI adoption is propelling Southeast Asia's digital economy to a projected $368 billion in 2026 and $1 trillion by 2035.”
Previous discussion: https://news.ycombinator.com/item?id=49977979 (2044 points 15 days ago, 1254 comments)
I think that this link is more helpful: https://mistral.ai/news/mistral-large-4/ Seems like it's multimodal and visual intelligence capabilities are better than Deepseek 4.1 Flash, even though coding is not equal to DS 4.1 1017. The cost i…
GGUF 형식으로 나와 있습니다.
“The country’s prime minister expressed disappointment at being informed of the hack only via email. Now Australia is investigating whether OpenAI broke the law.”