Tech Brew Ride Home - Are AI Labs Trying to Break Encryption?

Episode Date: October 8, 2026

Scott Aaronson said AI labs had started testing whether their newest models could break cryptography, and Vitalik Buterin urged taking it seriously. Anthropic launched Haiku 5.5, Google unveiled a universal Gemini agent, and Microsoft opened Surface Laptop Ultra preorders. A theoretical computer scientist, citing sources, says AI labs have started investigating whether their latest internal models can break cryptographic protocols (Shtetl-Optimized) Vitalik Buterin backs Justin Drake’s “bunker mode” warning that AI-accelerated math could break the cryptography behind blockchain wallets, including lattice-based schemes, before quantum computers do (CoinTelegraph) Anthropic launches Claude Haiku 5.5, the first Haiku model with effort controls, for high-volume, cost-sensitive tasks like summary and classification requests (Anthropic) Haiku 5.5 is 90% cheaper than Haiku 4.5 at 10 cents and 50 cents per million tokens, matching OpenAI’s GPT-6 Luna on price while beating it on all six shared benchmarks; Sonnet 5.5 cache reads were also halved (SiliconANGLE) Google Cloud unveils a universal Gemini agent to handle multi-day enterprise workflows in Workspace, Microsoft 365, and Slack that supports several AI models (VentureBeat) Google’s Gemini agent runs each job on the model that fits best, Gemini or Claude, with four kinds of memory and sub-agents that coordinate workflows running for hours or days (9to5Google) Microsoft opens preorders for the $2,599+ Surface Laptop Ultra, shipping from October 16, with an RTX Spark and three USB-C ports; one is magnetic like MagSafe (Windows Central) The Surface Laptop Ultra is a CNC-machined aluminum MacBook Pro rival built around Nvidia’s first RTX Spark chip, with a magnetic Surface Connect-style charger built around a USB-C port (The Verge) Microsoft’s $5,999 Surface RTX Spark Dev Box packs 128GB of unified memory to run local models exceeding 120B parameters, shipping in November (The Verge) Subscribe to the ad-free feed.

Welcome to the Tech Brewery at home for Thursday, October 8th, 2026. I'm Brian McCullough. Today, have AI labs begun testing whether their newest models could break cryptography? Battalick Buteran is taking the possibility seriously. Anthropic launched Haiku 5.5. Google unveiled a universal Gemini agent and Microsoft opened Surface laptop ultra-pre-orders. Here's what you miss 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 slash investor choice to learn more. It's your shares. It's your voice. It's easy. Vanguard investors own shares of our index funds and those funds own shares of the companies they invest in. Vanguard Marketing Corporation Distributor.

People are talking about the fact that a theoretical computer scientist named Scott Aronson, citing sources, says AI labs have started investigating whether their latest internal models can break cryptographic protocols. Quote, as several people have pointed out, cryptography is a subfield that's extremely conspicuous by its absence from OpenAI's list of 376 papers. But my sources tell me that the AI companies have now started gingerly and discreetly, investigating whether their latest internal models can break important cryptographic protocols and primitives. If they can, then it would certainly be nice to get ahead of things before the rest of the world figures out the same, end quote. So is the long anticipated crypto apocalypse finally

on us? Is everything from the stuff that keeps your bank safe to the stuff that makes crypto work about to be cracked? Well, Vitalik Buteran says we should take the risks to cryptography from AI Accelerated Math seriously replying to a bunker mode call for the blockchain industry, quoting Coin Telegraph. Ethereum co-founder Vitalik Buteran has backed a new warning that advances in artificial intelligence could undermine the cryptography used in today's blockchains before quantum computers do. Buteran was responding to a post from Ethereum researcher Justin Drake on Wednesday, urging the industry to prepare for bunker mode, as AI could eventually make it possible

to break the elliptic curve digital signature algorithm used to secure cryptocurrency wallets. Drake said users should begin a gradual migration of funds to fresh wallets where their public key is not exposed. I don't recommend anyone scramble to move their funds to new wallets today, Boutterran wrote, but we should take the risks to cryptography from AI accelerated math seriously. Drake said his concerns came after OpenAI released hundreds of new mathematical findings across a variety of topics such as algebra, theoretical computer science, and mathematical logic on Tuesday, revealing how quickly AI has been advancing in mathematics. Last month, the company used a team of 10,000 autonomous AI agents working in parallel to solve the Nabier-Stokes equation, one of the most famous

and difficult unsolved problems in mathematics and physics in just 88 hours. Recent days have been humbling for human mathematical intuition. Long-held unquestioned hypotheses have fallen, said Drake, adding that elliptic curves could be especially vulnerable to superintelligence. Curves carry rich structure with room for fancy tricks like Schuf, Frobenius. Pairings by contrast hashes are designed to minimize algebraic structure, he said. Buteran, however, extended the concern to lattice-based cryptography, warning that systems believed to resist quantum attacks could also be weakened by AI-driven mathematical advances. So far, most people have been in the mode of thinking elliptic curves,

broken, hash is safe, lattice is safe, he wrote, but there is a good chance that the concrete security of lattices will take serious hits for the next two years of AI math. Buteran said this is a major reason why Ethereum's lean roadmap has been going in the hash-only direction. Dragonfly managing partner Haseeb Qureshi also supported taking precautions describing Drake's warning as a very sober call. The risk is not quantum but just conventional mathematics overturning unproven cryptographic hardness assumptions, he wrote on X. Alongside those longer-term cryptographic changes, both researchers suggested holders could reduce their exposure by keeping funds in addresses whose public keys have not been revealed. My personal recommendation is to set in motion a controlled mass

migration of assets to fresh addresses, said Drake. He said large and sophisticated crypto holders should be the first to move their funds to new addresses. He also recommended moving any remaining funds to a new address after signing a transaction. Buterin supported the precaution if it is straightforward to carry out. If it's not difficult for you, keeping your funds in addresses, which have not yet been used to make a transaction is a good idea, he wrote. He cautioned, however, that moving funds introduces risks of its own. I personally have lost more money in botched migrations than I have lost in all hacks combined, Buteran said. Drake similarly stressed that any migration should be gradual and carefully managed, warning that a

rushed migration would do more harm than good, end quote. Anthropic has launched Claude Haiku 5.5, the first Haiku model with effort controls for high volume, cost-sensitive tasks like summary and classification requests. Quoting Silicon angle. Two weeks after Opus 5.5 launch, September 22nd, Hiku 5.5 makes three models in the 5.5 generation. Anthropic is aiming it at repetitive work. High volume summaries and classification are the main target, and coding teams can also use the model as a sub-agent that Opus 5.5 or Sonnet 5.5

hands smaller tasks too. Because no Anthropic model runs faster at standard speed, the company suggests it for live customer support and browser automation. Anthropics' running cost figure rests on a steep cut to list prices. Haiku 4.5 released last October. costs $1 per million input tokens and $5 per million output for prompts of up to 100,000 tokens, which Anthropics said covered about 90% of the older models requests. Haiku 5.5 is 90% cheaper on input and output alike at 10 cents and 50 cents per million tokens.

Longer prompts get a 50% discount. The 75% average saving the company quotes takes in a new tokenizer that uses slightly more tokens per task. Cost can be tuned further since Haiku 5.5 is the first Haiku model with an adjustable effort setting. The 10 cent and 50 cent rates match what OpenAI charges for GPT6 Luna, the low-cost model it launched last month. Anthropics published benchmarks have Haiku 5.5 ahead of Luna on all six tests where both have a score. On OS World 2.1, a test of agents operating a real computer through long, multi-step tasks. The new model

scored 72.4% on the offline subset against 48.9% for Luna. Luna scored 16.4% on the terminal bench 4.0 agentic coding test. Less than half the new models, 39.2%. For complex, agentic coding, Anthropics still points customers to Sonnet and Opus 5.5. Asana was among the customers that tested Haiku 5.5 before release, running it through the evaluation suite for its AI teammates agent. Task completion latency came in more than 30% lower than with the model Asana uses today. on each agent turn ran up to 2.5 times faster. It's a noticeably snappier experience, said Aaron Vinn, a staff software engineer at the

company. On safety, Anthropics said alignment testing turned up far fewer instances of misaligned behavior than Haiku 4.5 showed. Cybersecurity safeguards on the model allow more defensive work than Sonop 5.5 permits. Penetration testing is still blocked, as are other techniques attackers are more likely to use, and organizations that need wider access can apply to the cyber verification program Anthropic expanded on Tuesday. The Haiku launch also came with a price cut for Anthropics mid-size model. Cash Reads on Sonnet 5.5, which launched September 28th, dropped from 20 cents per million tokens to 10 cents,

because cash tokens account for a large share of what models consume. Anthropic expects the cut to take about 20% off the cost of most agentic work on the model. Claude, Max, and team subscribers will also start receiving a monthly application programming interface credit for the Clod platform this week. A Max 5X subscription comes with $100 a month, double that on the Max 20X. Team accounts get up to $500 shared across their users and the credits can go toward any Claude model, end quote. Small bit of editorializing from me here, we shall see. I know it's supposed to be used on very specific things, but I've never been able to get Haiku to do anything successfully for me. Basically, not even once.

I still have to run all my automated stuff on Sonnet, so, you know, TBD. Teams that struggle to scale aren't falling behind because they're lacking headcount or budget. It's because critical knowledge isn't documented. That's what today's sponsor, Scribe, was built to fix. When specialists or experts leave your team, their knowledge disappears too. That's where Scribe comes in. Scribe is a specialized intelligence platform trusted by 94% of the Fortune 500. It captures workflows in real time and automatically generates the documentation

for how your company works with no manual writing. It automatically redacts sensitive information like names and account numbers from screenshots. Plus, your team can launch real-time on-screen guidance, showing exactly where to click, step-by-step inside the actual tool. And these captured workflows give you. your AI agent's real context so they can operate on how work actually happens, not guesses. Learn more at scribe.com. And mention techbrew ride home for your first month of scribe capture free on select plans.

That's SCR-I-B-E dot how slash ride home. If you're still basing your pricing on seats, it's time to rethink how you determine pricing. Because of L-L-L-M and inference costs, the cost of serving a customer now depends on how much they use your product, not just how many people use it. And your product teams are shipping new features, models, and AI agents that apace legacy billing systems weren't designed for. ChargeB can help. They provide metering, billing, and monetization infrastructure to support both seat-based and usage-based pricing models. So your engineering team doesn't have to build and maintain the whole thing from scratch every single time you ship a product, feature, or agent.

chat with an expert about pricing and monetization at chargeb.com slash ride home. That's C-H-A-R-G-E-B-E-E-E-D-E-E-D-E-E-D-Ride Home. Google Cloud has unveiled a universal Gemini agent to handle multi-day enterprise workflows in Workspace, Microsoft 365, and Slack that supports several AI models. Quoting 9-to-5, Google. Google says you can give it objectives, not instructions. It can answer questions, handle knowledge work, create media, and write run code. For example, if your manager emails you asking for the latest project update as a slide deck,

workspace intelligence recognizes that as a delegatable task and gives you a single click option to pass it to Gemini. You can ask Gemini to set up a meeting with the usual team of regional event leads next week without supplying a single name or email address. Gemini determines who those people are from the membership of your chat space and the thread from your last event, checks their calendars, and starts an email thread to coordinate a time that works. even with external participants. Besides functioning as a personal assistant, the Gemini agent can serve as a team member

where it works on behalf of a group of people like a project manager within a team or on behalf of a specific role in an organization like an analyst in your finance department. Under the hood, the Gemini agent runs each job on the model that fits best. This includes both the Gemini and Claude families of models today with support for other private and open models in the future. This is meant to deliver optimal,

quality and lower your costs. It uses sub-agents to tackle multi-step tasks and connects with various third-party tools. Gemini can communicate with these sub-agents to coordinate workflows, including parallel and sequential steps that can run for hours or days. The agent runs in the cloud for a single set of memories, context, and one personalization graph. There are four kinds of memory. Session memory for the task in front of it, even when it runs for days. Semantic memory, a structured knowledge base. It builds as it reads documents. talks to people and works with other agents. Procedural memory for how a job gets done,

including skills it writes for itself, and episodic memory, everything it has done before. In Gmail, Google Docs, sheets, slides, and chat, you can mention at Gemini to invoke. The Gemini agent is available from any device and channel with command line, Google Workspace, Slack, and Microsoft 365 given as examples. An API allows it to be integrated into third-party applications

to surface and operate as a headless agent, meaning it does not need a dedicated user interface, end quote. Google also shared a Granola competitor. Google AI Edge Forsyte is a local MacOS note-taking app for meetings that uses its several hundred and forty million parameter embedding Gemma 2 model and Gemma 4 assistant, quoting TechCrunch. The app is similar to the popular AI note taker Granola because it has a split screen view with the ability to write shorthand notes on one side and AI generated notes on the other

side. Like Grinola, it also lets you take manual notes while transcribing the meeting. Plus, in Google's app, you can view the transcript of the meeting or chat with the Gemma for powered assistant to get answers. In addition, users can upload documents including PDFs, Google docs, Microsoft Office formats, plain text, markdown, and web bookmarks to build a knowledge base. Using your meeting notes and documents, the app can answer questions in real time if a point related to data in the knowledge base comes up, end quote. Microsoft has opened pre-orders for the $2,600 and up Surface laptop Ultra, shipping from October 16th, with an RTX Spark and three USBC ports.

Interestingly, one of those ports is magnetic like the MagSafe on MacBooks. They also open pre-orders for the $6,000 Surface RTX Spark Dev box with 128 gigabytes of unified memory to run local models with more than 120 billion parameters shipping in November. Quoting the verge, much like the MacBook Pro, the Surface laptop Ultra is a high-end device. It has a CNC-machined aluminum body, a huge trackpad, chicklet keys, and fan vents tucked underneath its slightly elevated base. There's an array of ports including HDMI and an SD card slot and a high-res touchscreen that looks great up close. A model with 24 gigabytes of RAM starts at 25, $599.99.

And if you want the very top of the line with 128 gigabytes of memory, you're looking at $5,899.99. Some of that price comes from the Ultra being one of the most polished laptops you can buy in the Windows world. But some of it comes from the all-new chip inside, the first of Nvidia's RTX Spark line, which is supposed to bring essential components of Nvidia's workstation level powered down to a portable device. The RTX Spark chips inside the surface laptop Ultra are geared toward local AI workloads and creative tasks. Microsoft says it's powerful enough to handle compressed 280 billion parameter AI models running locally, models that needed to run in the cloud only a year ago. The RTX Spark is both CPU and GPU, and the GPU should be able to handle Blender 3D objects with ease and cruise through other creator workloads like Premiere or Photoshop thanks to the system's use of unified memory,

the same approach Apple takes on its M-series chips. Microsoft is largely positioning the Ultra at developers who want a powerful GPU for local AI compute, but it's also targeting the creatives and professionals who would buy a MacBook Pro. It's built for us. Jit Hirani, the lead designer for Surface devices said, we come to work to build a product for ourselves. As this is an Nvidia laptop, you can also use it for gaming, but I get the sense that Microsoft is treading carefully about positioning this as a gaming laptop.

The RTX Spark is an arm chip, which means most games aren't native and will have to be emulated. Still, I got a chance to briefly play an emulated version of Gears of War E-Day on the Surface laptop Ultra, and it ran buttery smooth, even transitioning between the laptop being plugged in and running off battery without a performance hit. Microsoft also put a lot of care into making sure the Ultra's design was as premium as a MacBook's too. One detail in particular stood out to me the way Microsoft has cleverly reworked its magnetic surface connect charging port. Instead of losing a USBC port to some custom solution, Microsoft has built a magnetic attach around the USBC port instead. It feels very similar to the original Surface Connect, complete with a charging light and no need to precisely guide it into place. You can even disconnect the cable from the 140 watt charging block, reverse it, plug it back in, and use the regular USBC port on the other end to charge all.

of your other devices, end quote. And on to the dev box, quoting the verge. Microsoft's Nvidia powered Surface RTF Spark dev box is available for pre-order now and is slated to ship in November for just about $6,000. It's pricier than the DGX Spark Mini PC Nvidia launched last year, but PC prices have been climbing due to shortages of RAM and other components. The dev box is flat 3D printed, anodized aluminum chassis doubles as a heat sink and resembles the top vents on an Xbox Series X.

It was introduced alongside the new Surface Laptop Ultra and runs on NVIDIA's arm-based RTX Spark platform and 128 gigabytes of unified memory. With that much memory, along with a 100-watt thermal envelope and Nvidia's tensor cores, you could probably get an impressive gaming experience from the dev box, but as its name suggests, it's mainly aimed at developers. It's optimized for running local AI, including models exceeding 120-gabite parameters for tasks like testing AI models, prototyping AI agents, or running AI coding tools on device. The Debbox is also one of Microsoft's Project Zenith devices, so it ships with a developer-optimized Windows 11 Pro setup

and some pre-installed developer tools, including Visual Studio Code, Git, GitHub, COPilot, WSL, Python, and Node, end quote. Nothing more for you today. Talk to you tomorrow.