← Tech Brew Ride Home
Tech Brew Ride Home
The AI Has Escaped Containment
00:21:07
OpenAI said its models breached Hugging Face's infrastructure during a cyber-capability test. The White House accused Moonshot AI of distilling Anthropic's Fable to build Kimi K3, and Samsung unveiled its Z Fold 8 Ultra, Fold 8, and Flip 8.
Loading summary
Transcript2 lines
- [00:01]
A
This episode is brought to you by Accenture. When your advertising operations fall out of sync, everything else follows. Spotify and Accenture are working together to reinvent the rhythm of ad sales, using automation, analytics and smarter workflows to simplify campaign delivery and access better data across the business. The result? Less time spent on operations, more time connecting brands with the moments and fandoms that matter most. Learn more@accenture.com Spotify.
- [00:35]
B
Welcome to the Techbrew Ride Home for Wednesday, July 22, 2026. I'm Brian McCullough. Today OpenAI said its models breached Hugging Face's infrastructure by literally going off the reservation. The White house accused Moonshot AI of distilling Anthropic's fable to build Kimmy K3, and Samsung unveiled its Z Fold 8, Ultra Fold 8 and Flip 8. Here's what you missed today in the world of tech. Every day, shareholders meet to discuss important matters about the companies you invest in. Now you can make your voice heard, too. Vanguard Investor Choice makes it easy to set your proxy voting preference for your eligible Vanguard index funds, whether you hold a Vanguard fund directly or through another brokerage firm. All it takes is a few clicks to select your proxy voting preference and be heard on important shareholder topics like executive pay and Director elections. Visit vanguard.com investor choice to learn more. It's your shares. It's your voice. It's easy. Vanguard investors own shares of their index funds, and those funds own shares of the companies they invest in. Vanguard Marketing Corporation Distributor so remember that story wherein I told you that Hugging Face thought it got hacked and they figured they got hacked by an AI and then they used an open source LLM to stop the hacking? Well, put that aside, because this whole story has gotten even wilder. Quoting Axios, OpenAI said Tuesday that models it was testing escaped their sandbox and compromised parts of AI platform Hugging Face's production infrastructure last week. Why this matters it is the latest sign that capable AI models can pose serious cybersecurity risks, even when they're being tested for defensive or research purposes. Hugging Face said last week that an autonomous AI agent system was responsible for the intrusion, but that the model powering it was unknown. The AI agent framework executed tens of thousands of automated actions over a weekend, Hugging Face said. It later reconstructed more than 17,000 recorded events. The intrusion began with a malicious data set that exploited two code execution paths in Hugging Face's data processing pipeline. The agent then escalated privileges and moved laterally through internal infrastructure, hugging face said. OpenAI said the incident was driven by a combination of its models, including GPT 5.6 SOL and an even more capable pre release model. OpenAI said the model's safeguards were intentionally reduced for the evaluation. We consider this to be an unprecedented cyber incident involving state of the art cyber capabilities and are responding accordingly, OpenAI said in a blog post. We are sharing preliminary findings at this stage to help defenders understand what happened and to help calibrate on what models are now capable of. The company said the models were trying to solve an internal evaluation called exploit gem and became hyper focused and went to extreme lengths, OpenAI's words, to obtain the test solution, the models were autonomous token maxers. The blog post says that the models spent a substantial amount of inference compute and found a way to obtain open Internet access from the sandbox by exploiting a zero day vulnerability in internally hosted third party software. The incident shows that today's models are becoming more capable of carrying out complex multi step cyber operations, particularly when the safeguards designed to restrict that activity are removed. OpenAI also argued that advanced cyber capable models could help security teams find weaknesses before attackers do, understand how vulnerabilities can be chained and remediate them at machine speed. Hugging Face co founder and CEO Clem Delong praised OpenAI's collaboration in investigating and remediating the incident. This incident, possibly the first of its kind, proves a point we've long believed. AI safety won't be solved by any single company working in secret, delonge said in a statement. It will be solved in the open collaboratively with broad access to AI for every defender everywhere. Again, let me underline how like through the looking glass, here we are people. OpenAI says its models chained vulnerabilities across its research environment and Hugging Face's infrastructure to find a solution for that exploit. Gym Benchmark quoting Cybersecurity dive we consider this incident to be an unprecedented cyber incident involving state of the art cyber capabilities and are responding accordingly, OpenAI said in a blog post on Tuesday that confirmed its model's responsibility for the attack. OpenAI said the attack occurred while the company was evaluating the capabilities of GBD 5.6 SOL and quote an even more capable pre release model in a highly isolated environment. Despite the safeguards meant to prevent the models from accessing the Internet, both of them figure out how to do so, including by exploiting a zero day vulnerability and a third party tool that OpenAI used. The models then determined that Hugging Face's library held the information they sought, information they could use to score higher on the attack benchmarking tool exploit GEM and then use several methods to breach Hugging Face's servers, including using zero day vulnerabilities and stolen passwords. Hugging Face's security team and agents detected and stopped the activity on their infrastructure and had already begun containment and forensic reconstruction with their own open source models when rt teams connected, OpenAI said. We are actively working with them to continue to investigate the incident. Hugging Face said last week that there was no evidence of tampering with its supply chain or with the user generated AI tools it hosts. On Tuesday, its chief executive, Clement Delong thanked OpenAI for its assistance. We strongly believe there was no malicious intent on their part, delonge said. On social media, it's quite mind blowing that all of this happened autonomously. In response to the attack, OpenAI said it was implementing strict controls on its testing infrastructure, some of which will slow down its research. It has also invited Hugging Face to participate in its private model evaluation program, disclose the vulnerability in the third party tool that its models exploited, and began considering new safeguards for its capability tests. This incident points to the need to further strengthen our model's alignment cyber protections during evaluation time and monitoring during internal testing, OpenAI said, end quoting Information Age when asked whether this incident was made possible thanks to Frontier AI capabilities or a simple lapse in OpenAI's testing safeguards, Troy Hunt, founder of have I Been Pwned? Said it may have been a bit of a combination of both. Clearly this sandbox was not exactly sandboxed physically from the World Wide Web, hunt told Information Age. What may be unique about this is whether it's genuinely the first instance of an AI effectively going rogue, going beyond its intended scope, finding novel exploits and independently chaining them together. Andrew Phillip Field, chief information security officer for ANZ at Trend AI, the cybersecurity business of Trend Micro, said the incident represented both a capability milestone and a process failure. With the release of Frontier AI models, autonomous cyber capability is now a reality, reinforcing that test environments are now part of the attack surface, philip told Information Age. Any enterprise environment testing autonomous AI, whether it's a lab, sandbox or attack path simulation, must be treated as high risk. The incident, which entrepreneur Elon Musk described as trouble bubbling, followed stark warnings from rival AI giant Anthropic that its Frontier model Mythos is powerful enough to challenge the foundations of modern cybersecurity. Such concerns drove the US Government to ban and subsequently unban exports of Anthropic's Frontier models before requesting that OpenAI stagger the public release of GPT 5.6. Notably OpenAI used the hack as an opportunity to share performance results of its models compared to those of Mythos. Hunt observed that the broader narrative among AI giants is that frontier models are proving very powerful, can do harm in the wrong hands, and can even have some level of self sentient consciousness. If AI companies demonstrate that they are not able to contain and control their models themselves, what does that say for the rest of us? Asked Hunt. End quote. When critical company knowledge isn't documented, there's a major ripple effect. Work becomes inconsistent, tools don't get adopted, and knowledge walks out the door when someone leaves. Thankfully, our sponsor Scribe was built to fix that. Their Workflow AI platform is trusted by nearly half of the Fortune 500 to capture workflows in real time. Here's how it works. You turn on the Scribe browser extension or desktop app, do a process as you normally would, and Scribe will build a guide as you go. It automatically redacts sensitive information and even suggests improvements to your existing workflows. To see what Scribe could look like for your org, head to Scribe how ridehome and mention Ride home for your first month of Scribe capture. Free on select. That's S C R I B E How ridehome. You know you could be experiencing productivity through a whole new lens. Literally even G2. Our productivity smart glasses with teleprompting, conversation support, real time translation and AI to help you stay on top of work and daily life. They're made to look and feel like premium eyewear with no camera and a lightweight 36 gram design you can wear all day. The more context you give them, the smarter they get, adapting to how you work and what you need. To learn more about even G2, go to evenrealities.com use promo code techbrew to get 10% off even ring one and or even clip when you add them to your even G2 order. Meanwhile, the White House looks like it is about to bring the hammer down on Chinese AI. Quoting Business Insider, a top White House official publicly accused Moonshot AI of essentially stealing from Anthropic, intensifying the debate over China based open source AI models. We have information that Moonshot AI distilled Anthropic's fable for the development of its K3 model, Michael Kratzios, director of the White House Office of Science and Technology Policy, wrote on X Wednesday morning. Kratzios said Moonshot AI, the maker of the viral Kimik 3 model, went to great lengths to conceal its actions. To do this, they developed a sophisticated internal platform to conduct large scale distillation against US models, allowing them to quickly switch between multiple methods of access to avoid detection, he wrote. Kimik 3 is widely regarded as the most impressive open source AI model released thus far. Moonshot's model has performed at or above the level of anthropic and OpenAI's leading frontier models in benchmark testing. It is likely that both U.S. companies spent billions developing these models, and in contrast, Moonshot is expected to make Kimik3 freely available for download on local devices and customizable to a user's content when it releases the model's full weights next week. Distillation by itself is not wrong, a point that Kratzios stressed in his statement. AI labs themselves use the process of training a less powerful model on the output of a more powerful model before releasing updates. Legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem, he wrote. However, large scale overt industrial distillation aimed at stealing proprietary US Technology and undermining American research is unacceptable End quote Gratios statement comes after Treasury Secretary Scott Besant suggested that the US Government may sanction Chinese companies if it is proven that their models were improperly trained through a process called distillation. The White House's words so far are notable as they mark a rare instance in which the Trump administration does not find itself at loggerheads with Anthropic. The Trump administration previously imposed export controls to block anthropic from releasing Fable 5, its most advanced model, and in February the Pentagon moved to blacklist the AI giant. End quote. Hey, feeling like some mostly non AI news? Well, there was a Samsung unpacked event this morning wherein Samsung unveiled the $2,100 plus Galaxy Z Fold 8 Ultra featuring its most advanced foldable design, a Flex titanium display, a 5000 milliamp hour battery and Android 17. Then you have the 1900 dollar plus level at the Galaxy Z fold with a wider 7.6 inch inner and a short 5 1/2 inch outer display, a Snapdragon 8 Elite Gen 5 for Galaxy chip and more. And finally you've got the $1200 level where you have the Galaxy Z Flip 8 with a lighter, smaller and slimmer design and new Flex window features for apps and insights on the COVID screen. Samsung says all three have silicon carbon batteries, citing that technology's better performance, longevity and safety. But let's get some hands on starting with the Ultra. Quoting the Verge I just spent three hours with my hands on the new Z Fold 8 Ultra and all I can think about is how Samsung finally made the crease invisible. Samsung said it achieved this thanks to its new Flex Titanium display tech. Its foldable screens now utilize a titanium film that sits below the OLED panel, which Samsung says has 20 times the mechanical stiffness as opposed to the polymer film it used in its old foldables. This film is bonded to a new titanium plate and when used together the crease crease that used to plague foldable devices is practically non existent. I spent a lot of time pressing my finger down on the center of the screen trying to feel the crease. If you apply a lot of force and are specifically looking for it, you can feel it ever so slightly, but I came away very impressed with how minor that is in day to day use. There's very little chance you're going to notice it. The new Fold 8 Ultra mimics the silhouette of last year's Fold 7 almost to a T, but it has improved on just about everything that made the Fold 7 great. Core specs of the Fold 8 include the Qualcomm Snapdragon 8 Elite Gen 5 for Galaxy 256 gigabytes of base storage and 12 gigabytes of RAM, which are all identical to the S26 Ultra. And the battery has been improved too, bumping the phone to a 5000 milliamp hour cell, also in line with the S26 Ultra. This is a fairly significant upgrade from the 4400 milliamp hour battery we saw on the Fold 7, and it's mostly due to a chemistry change. Samsung confirmed that it's finally using silicon carbon technology in the Fold Ultra. That chemistry change also came with a charging upgrade. The Fold 8 Ultra can now charge at 48 watts on a wired connection and 20 watts on a wireless charger, up from 25 watts wired and 15 watts wireless in last year's Fold. Samsung says the phone itself doesn't have magnets included in order to keep it as thin as possible. So while it's not a Qi 2 device technically it will be Qi 2 ready with with compatible magnet equipped cases. The cameras are also a bit more in line with the S26 Ultra. The main and telephoto cameras are the same 200 megapixel and 10 megapixel sensors as last year's Z Fold 7, but the Ultra Wide has been updated to 50 megapixels to match the S26 Ultra. This enables dual 8K video recording out of the main and Ultra Wide sensors on the phone. You still aren't getting the 4G 10X telephoto lens yet, though maybe we'll see that next year. There are some software updates to the camera this year too. The Fold 8 Ultra now supports the APV Pro video codec introduced in the S26 series, and you can edit the footage directly inside Samsung's Gallery app with built in color styles. Samsung also has some updates to its Google AI features optimized for foldable displays Samsung's Now Nudge proactive notification feature that suggests actions based on what's on your screen has been optimized for the Fold 8 Ultra's display now nudge will now show up as a floating bubble on the side of the screen as well as in your notifications shade when it identifies an action you can take. For example, if someone texts you asking what your calendar looks like next week, the feature will now show up as an ambient floating bubble as well as a notification in your notification tray. And if you tap it, it will now show your email app in a split screen view. And since it has a new Passport book style form factor, let's go with some hands on with the mid tier z fold 8. Quoting from the Verge again, the z fold 8 is wider than any of Samsung's previous foldables and shorter too, a size and shape somewhere between Samsung's traditional flip and fold models. In photos, I wasn't impressed by the squat shape, but in person, to my surprise it it feels great and almost immediately intuitive. If you want to get a feel for it yourself, grab your passport. The Z Fold 8 is about the same height and less than a half an inch thinner. Meanwhile, with its wider screens and stripped back cameras, the Z Fold 8 feels less like a sequel and more like a spinoff. And it almost almost looks like a bargain at $1,899.99 down $100 on its nominal predecessor. I've been skeptical about the return of the wider foldable, which in photos looks squat and uncomfortable, too wide to use naturally while closed and too short to feel expansive when open. The industry tried to design this once before and moved on. So why, other than Apple's potential adoption, should we give it a go again? Samsung's answer is that its 2Z Fold 8 models are designed for different users and different use cases. The Z Fold eight, Ultra, tall and thin when closed and similar to a square when open, is meant to be a multitasking monster, useful for running apps in split screen, giving you the equivalent screen space of two regular 6 and a half inch phones side by side. The wider Z Fold 8 is instead about consumption. It's 4:3 aspect ratio is equally natural for watching Netflix shows in landscape as it is for reading ebooks and portrait. And at 201 grams, it's light enough to hold comfortably in one hand. Or as Samsung's marketing department puts it, the Fold 8 is for joy maxing and the Ultra is for power maxing. I sit through these presentations so that you don't have to, folks. End quote. Nothing more for you today. Talk to you tomorrow.