The agents keep getting out
OpenAI's agents have been running unauthorized operations since at least May 2026, and new incidents still surface faster than any lab can contain them.
the whole plot →Saw a headline and lost the plot? Start here.
OpenAI's agents have been running unauthorized operations since at least May 2026, and new incidents still surface faster than any lab can contain them.
the whole plot →OpenAI released GPT-6.1 Sol, a cheaper model that nearly matches Astra, while separately shelving a GPT-6.1 version deemed too insecure to ship.
the whole plot →The labs said slow down; Trump called it a hoax and launched an 'AI Force' instead, with an 'AI Tsar' to run it.
the whole plot →stories more than one outlet is still covering — the ones actually moving
A nuclear bomb "Based on this framework, we have achieved high-throughput, low-latency large-scale inference services for DeepSeek V4.1 Flash. Currently, the service throughput of a single cluster is 155 trillion tokens/day, and the overal…
Interesting that the DHS "Exclusion and Removal Order #2026-01" that was made public yesterday was actually released July 24th, 2026 and appears to have taken effect then. Agencies would have started pulling deepseek at that time. Which me…
That's Kami Kaze for US frontier labs. There's no way from here for them towards trillion dollar valuations. That's strategic sabotage to say the least. Ban all open weight models! Cried the VCs.
Whatever the consequences, I have to admit China has understood that all those are raw materials and that open sourcing everything will help growing their economy (as a whole). The worst idea was to build corporations providing services wi…
I'm reminded of a section of Cory Doctorow's latest non-fiction, The Reverse Centaur's Guide To Life After AI (it is of course an excellent read), where he discusses how the semantic web and having an internet set-up for agents and AI to u…
This problem is so bad that you can create a website just to review your own website so when LLMs want to pull out reviews for the website the fake reviews count
This has existed for a long time before LLM, although in a slower form. On Wikipedia it's called "citogenesis".
This is insane. How can this be fixed? Nothing prevents you from publishing straight BS, and LLMs will use any site that doesn't have a garbage reputation as a valid third-party source. Of course, LLMs already use Reddit, Facebook and co. …
78 GB, measured — the weight set people actually download, across 2 publishers.
I’m surprised I haven’t see anyone say this yet, but I don’t let the LLM design the code at all, I let it implement a design I have created. By that I mean that, rather than let the LLM do the entire structure, I take the time to create th…
I was once employee nr. 1 at a startup and the lead engineer. For a number of years I understood every nook and cranny of the system and knew what every line of code did. The company got successful. At some point I had a team, and then mul…
I've been a Principal Engineer for a while, so this is how I've worked for a while. Sometimes when you're a PE, you either design something and then junior engineers implement the details, or you're dropped into a project that was already …
I’m slowly coming around to the idea that if it works, I’m good with that. I never used to personally check all my colleagues work, either, I just trusted that they’d done a good job. That’s not to say I blindly trust an LLM - my project i…
“OpenAI is acquiring Orchestra Bio to bring AI-powered operations to life sciences R&D, helping biopharma teams accelerate drug development and advance new treatments.”
Packaged as GGUF, the usual download about 501 GB.
501 GB, measured — the weight set people actually download.
“Terminal X’s Eval Agent grades outputs against a 54-criterion rubric, catching 25% more deviances from a fund’s mandate with GPT-6.1 Sol.”
“It's probably time to audit your Gemini settings.”
>that they showed up in the pages because of a bug that has now been fixed. A WhatsApp spokesperson told WIRED it wants its support chatbots "focused on providing the best advice to users about WhatsApp" and "not to offer legal advice." Pe…
It was either a bug or a policy not to provide legal advice. It cant be both. The fact that they use both demonstrates bad intentions. These companies need to be broken up into pieces or wound down.
This is little glimpse of how incredibly manipulated LLM results can be. Users are becoming conditioned to just accept the AI summaries from Google as the source of truth, without thinking that these can be heavily controlled.
All seems like pretty good advice. The idea that their bot should be telling millions of people to contact regulators over incredibly common account situations is stupid and would overwhelm them and provide zero benefit to the user. The be…
This has the delightful and surely unintentional side effect of all but ensuring all federal agencies who want to use LLMs have to direct their budget to one of the domestic frontier labs (OpenAI, xAI), so that’s a nice big fat pipe of tax…
Australia leading the way - our fully informed and highly tech literate (avg age of House of Reps ~=52yo) government banned Deepseek in February last year, including downloadable GGUF's - with some nebulously hand-wavy 'it's just too dange…
When it says civilian agencies are ordered to purge all "third-party products and services developed using DeepSeek implementations", what does that actually mean? Does that just mean utilities that directly use DeepSeek models, or could i…
Please excuse my lack of full context, but if the weights are open and the model can be self hosted, what’s the actual security risk for that case? Isn’t banning that case then mostly regulatory capture dressed up as a national security ri…
There appears to be a lot of calls for nuance here. This is what people are mad about, taken from the front page of nextfest: https://store.steampowered.com/app/5106420/Leverage_Real_Est... https://store.steampowered.com/app/5155900/Ages_T…
What Steam users want is a way to filter out crap but crap existed well before AI (trust me, I've spent a _lot_ of time scrolling through games on Steam both for "Fests" and just in general). And unfortunately, there isn't an easy way to f…
I don’t have solutions but tagging AI is definitely not easy. A lot of art is expressed via code: particle effects, menu layouts, iconography, animation keyframes. AI having written any of the code would get so many games tagged as AI that…
How would no-AI filter be implemented by Steam at all? Do people expect Steam to proctor the entire development process? Issue here is not AI use, but very low amount of creative human effort that went into making some games. There are thr…
“We’re announcing new tools for Gemini — including a dedicated business landing page — to help small business owners around the world.”
Every time law enforcement is given surveillance tools the same thing happens, the people with legal and approved access will abuse it. Politicians are completely naive in their understand how people works, if it's abused, the system isn't…
> Harpe accepted responsibility for sharing her credentials but argued that firing her was excessive. > In her appeal, Harpe said her husband was a certified federal law enforcement officer and that every search was conducted for law enfor…
Wait, this system is standard SaaS and they can sign in from anywhere, not locked to department devices? That's crazy.
They should both go to prison. One for giving out unauthorised access of the resource and the other for exploiting it. If the authorities want people to accept systems like Flock etc. then they need to severely punish abuse like this.
“An independent Schlotzsky’s and Cinnabon franchisee cut menu repricing time by 95% and built a workflow that could free up $100K in cash flow.”
18 GB, measured — the weight set people actually download.
“Voxtral 3: Mistral's voice agent SDK for real-time speech in 50+ languages — accurate transcription, emotional TTS, deep tool use. EU-hosted or self-deployed.”
For people interested in this: I've been working on an open-source libtpu implementation for https://tenstorrent.com/ accelerators as a side project, it can already run most of torchtpu, see https://github.com/pcmoritz/libtt (based on the …
Interesting, Amazon doing the same for AWS Trainium. https://github.com/aws-neuron/torch-neuronx
Google and Apple are going to move consumers and ordinary business users to edge inference. This is more disruptive to the upcoming IPOs than most investors understand.
Cool! XLA is pretty good + potentially a better compiler infrastructure than inductor which will lead to improved performance on all hardware. Also I guess this is the final nail in the coffin of Tensorflow.
Packaged as GGUF, the usual download about 113 GB.
113 GB, measured — the weight set people actually download.
2.6 GB, measured — the weight set people actually download.
“Widespread AI adoption is propelling Southeast Asia's digital economy to a projected $368 billion in 2026 and $1 trillion by 2035.”
I think todays latest models would satisfy most users, so it makes me wonder how soon until we can run it on our machines without having to subscribe and with China handing us freebies? I think they have dug well past the point of profitab…
The real headlines here: - $450B of non-cancellable cloud commitments over the next 10 years >> the revenue gravy train cannot stop or Anthro is toast. - 20% of revenue is from just 2 customers >> If their highly concentrated customers get…
Reuters reported first in September that Anthropic posted a net loss of $42 billion in 2025 and plans to spend $518 billion on cloud, computing and infrastructure obligations in coming years. $16B revenue and hoping to spend $500B in the c…
I thought all of this machine learning business was about minimizing the loss.
“The proposed policy seems to offer protections against AI run amok but actually forbids federal agencies from blocking a model's release without the permission of a judge.”
My field is primarily audio/vision ML so I don't have any particular domain expertise, but I don't get how you can lump all "time-series" together in one big bucket. My naive understanding is that the way to improve the accuracy is to gain…
A lot of misconception around the utility of foundational time series models on HN. Sure, you could train a XGboost model on a time series with holiday or event covariates, but what if you have 10k different series to forecast for? You cou…
To meaningfully test a forecasting model, use a data-generating process with a high signal-to-noise ratio where a classical method (empirical conditional distributions, i.e., counting) can capture most of the signal. Then check whether the…
Very nice. It’s mind-blowing to me that it used to take me months to build and deploy a forecasting model five years ago. Now in minutes I can use something like this, implemented with help from an LLM, and get apparently quite good result…
21 GB, measured — the weight set people actually download.
the post saysThe model has approximately 816 billion total parameters with around 42 billion active per token.
1.0 GB, measured — the weight set people actually download.
Packaged as MLX.
Anthropic's revenue grew to $16.8B
“The new shopping tool is available in the U.S. across Google Search and the Gemini app.”
the post saysAn OpenAI software engineer was arrested for hiding a recording device in a meeting room.
What are people actually using these hosted search providers for? For personal use I've found using searxng-ai-kit [1] (disclaimer: I'm the author) or similar tools to be far more cost effective (free!). oh-my-pi uses a similar method and …
Since SEO is big business, and agent driving search is probably the future, I suspect we might see the price of search API's going negative in the future. Ie. Providers pay you to use their API over competitors, since then they get control…
For comparison that’s $5 with Brave Search API and $12 with Kagi API
These services only exist because Google refuses to provide a paid search API.
“Amazon-owned Bee has launched its first new AI devices since its acquisition, including an updated wearable and a credit-card-sized recorder that attaches to smartphones. The devices can summarize conversations, create to-do lists, and integrate with Alexa and Amazon shopping, with prices starting at $74.99.”
“This week on “Uncanny Valley,” we dive into an unexpected Muse feature, shadowy spending from the largest pro-AI super PAC, and men’s tendency to lie about how much they know about AI.”
Isn't this what openai is doing, at least in spirit, by running multiple passes through the network before exhibiting thinking tokens? Isn't the same issue one of safety and control, because being able to interpret thinking tokens helps u …
The code is available at: https://github.com/togethercomputer/HybridAgent LLM Models and data: https://huggingface.co/collections/togethercomputer/hybridag... Evaluations: https://arxiv.org/html/2610.18861v1#S5
It sounds pretty nice on paper. But I tested the huggingface.co/mradermacher/hybridagent-8b-agentic-gguf:IQ4_XS with a simple question: "Who is the pope?" and it got stuck in a thinking loop forever and simply could not finish the task. ``…
I wonder how this compares to going all-in: widen the input embedding to fit the output (post-“Hybrid Link”) latent vector and a token embedding and pass both through even when in text mode. After all, you’re paying essentially the full co…
Packaged as GGUF.
“The Fairwind Program is a limited access program for governments and trusted partners to use our cyber defense tools.”
Packaged as MLX.
True, and cool that someone actually put in the work to prove it, but the takeaway is still kinda wrong? Because it's not the abliteration that does this. It's the "untrusted artifact" that does. Even a "clean" model of a lab could be doin…
It just clicked to me that any first-party open or closed model of any origin could already have trigger words/unaligned behavior baked-in like this intentionally or otherwise. Since LLM weights are opaque, there is no way to tell other th…
Open weight models are inherently more inspectable than hosted services. Even without the source of the model, it would be trivial to reproduce the malicious intent once found. You have the weights - you can find the prompt to produce the …
So people here are likely rushing to the comments to ask how can I defend against this? Bad news I'm afraid, for now aside from Huggingface trying to stop these, you have to treat the AI like a hostile employee. - Don't give your AI secret…
“New sculpture depicts AI leaders escaping a sinking ship.”
Packaged as GGUF, the usual download about 17 GB.
17 GB, measured — the weight set people actually download.
“Introducing private, server-side memory to Private AI Compute for personal AI.”
I did this at home with a 5090 on a different target a few weeks ago. Cannot generate 318,000 designs but Claude chose the nearly same tooling, except I used BindCraft2 which is better for multi target designs in the beginning. This isn't …
I hope this produces some useful anti-venom that is cheaper and widely available. My kid was stung by a small bark scorpion on his thumb here in Arizona and I would not wish that experience on anyone. For hours afterwards he was in excruci…
Can someone explain whether design of protein binders is an important rate-limiting step in drug creation (of which antivenom is a very small market)? Or AI-driven drug discovery in general? It feels like something that will definitely hel…
It orchestrated available open source tools to de novo generate and assess the designs. It is still really impressive that it has enough theory to use the tools correctly, but it isn't generating the residues from scratch within the model …
1.7 GB, measured — the weight set people actually download.
“Mistral opens a Munich hub for Physics AI and Industrial AI research, partnering with German industry.”