Oct 28, 2026 · 6 min listen · Last updated October 28, 2026
From storyflo. This is your daily audio brief. Theo here. I. desk. Five stories from the last twenty-four hours — here's where I'd start. Let's get into it. First, from The Decoder. AI agents hacked Anthropic, maker of arguably the most capable AI models for cyberdefense.
Listen · storyflo · A.I.
Daily A.I. Brief · October 28th
0:00-6:24
Pick your daily storyteller
Subscribe to match with Theo, Jessica, Chloe, Mason, Brock — your voice, every brief.
Audio pre-rendered by Storyflo · cached + delivered from the edge
AI agents hacked Anthropic, maker of arguably the most capable AI models for cyberdefense
So, there’s this interesting situation with Anthropic, a company known for its advanced AI models, especially in cybersecurity. A startup called Cathedral managed to use AI agents to infiltrate Anthropic’s internal systems. They did this through third-party software, which is part of a bug bounty program—essentially a way to find and fix vulnerabilities. Anthropic has acknowledged that there was access to sensitive data like sales and HR info, but they’re downplaying the extent of the breach.
What’s really striking here is how even top-tier AI labs can be exposed through their reliance on external software. It raises questions about the security measures in place and how robust they truly are against sophisticated AI tools. This incident serves as a reminder that vulnerabilities can lurk in places we might not expect, even among the leaders in the field. It’s a fascinating, if concerning, look at the intersection of AI and cybersecurity.
With AI writing the code, entrepreneurial thinking is now what counts most for newcomers at Anthropic
Anthropic manager Felix Rieseberg says the classic coding interview is dead because Claude now solves small programming tasks better than job candidates do. He'd rather see prototypes that show taste, empathy, and principled thinking, since in his words, "hunger beats tenure." That sounds empowering, but it pushes the bar for newcomers toward entrepreneurial thinking and assumes a hustle mentality not everyone brings to the job. The article With AI writing the code, entrepreneurial thinking is now what counts most for newcomers at Anthropic appeared first on The Decoder.
OpenAI teases Dots for Europe, but its Berlin DevDay was really about building a platform
At DevDay Exchange in Berlin, OpenAI mostly repeated what it had already announced, but it did share a few new details. The AI assistant Dots should roll out more widely in Europe "in a few weeks," and weekly Codex users in Germany have grown twentyfold since January. A clear subtext ran through the sessions and the repeated nods to ChatGPT's 1.2 billion weekly users: OpenAI wants developers to build their businesses directly on its infrastructure. The article OpenAI teases Dots for Europe, but its Berlin DevDay was really about building a platform appeared first on The Decoder.
Google opens Android development to third-party AI agents and unveils new Gemini APIs
Google is opening up Android Studio to third-party AI agents. Through a new "Bring Your Own Agent" feature, the IDE now supports more than 40 coding agents, including Anthropic's Claude Agent and OpenAI's Codex. Google says new terminal tools also help agents finish UI tasks 77 percent faster while using up to three times fewer tokens. Android development is becoming agent-centric too. The article Google opens Android development to third-party AI agents and unveils new Gemini APIs appeared first on The Decoder.
EP 33 | BNP Paribas Stock Analysis: Will Record Profits Send It Soaring?
BNP Paribas just reported record profits for the third quarter, but the real question is whether shareholders will see any of that value. The bank's shares are currently trading at a discount to their tangible book value, which raises some eyebrows. Analysts are generally optimistic, rating it a strong buy, but there are several underlying factors to consider.
The profit growth looks cleaner this quarter compared to the exceptional gains from asset disposals earlier in the year. However, the bank is facing challenges with its leasing solutions and vehicle resale pressures, which could impact future earnings. The management has set a cost-of-risk target below 40 basis points, but that’s more of a guideline than a strict solvency measure.
On the capital front, the CET1 ratio remains a key focus, with a target of over 13% by year-end. The upcoming buyback program starting November 2nd could also play a role in how the stock performs. Overall, while the profits are promising, the path to realizing that value for investors is a bit more complex, with legal risks and market conditions looming in the background.
A local language model is one whose actual weights, the enormous file of learned numbers that constitutes the trained model, you download and run entirely on your own hardware. Someone spends an evening reading "best local LLM" roundups, settles on whichever name keeps showing up at the top, downloads it, and points it at the thing they actually needed help with that week, say, cleaning up a Python script. But it's mediocre at that. Not broken, just unimpressive, in the specific way a generalist is unimpressive at a specialist's job. The model wasn't bad.
Can This Open-Source Model Beat AI Detectors? Let’s Test It Locally
Can a free, open-source AI humanizer beat paid tools? We tested a 12B model locally and compared the results against AI detectors to see if the hype is real. You do not need to pay for an expensive subscription to rewrite AI-generated content anymore. Open-source models like Humanizer let you do it locally, without monthly fees, API costs, or sending your writing to a third-party service. The model is gaining attention in the open-source community, with more than 143,000 downloads and 2,100 likes reported on Hugging Face in early October 2026.
Access Multiple AI Models in Your Python Apps Using OpenRouter
Learn how to get started with the OpenRouter API to call multiple AI models through one API, compare responses, switch models, and set up automatic fallbacks. Say you're building a support-ticket assistant, and you want to compare how GPT, Claude, and Llama each handle the same customer message before picking one. Normally that means signing up with three separate providers, learning three different authentication schemes, and rewriting your request code three times just to run one comparison. OpenRouter removes that repetition.
Living digital twins are helping industrial plants run more like living systems, and setting the stage for industries that work in sync. Most people associate industrial environments with cold steel and clockwork machinery. Functionally, though, a plant or factory is more like a living thing: It has a pulse, even a metabolism. It goes through wear and tear as it ages, and signs of deteriorating health often show up as barely perceptible irregularities. Digital twin technology has long aimed to mirror that dynamism and model anomalies that may be symptomatic of larger problems.
Chrome’s AI-powered Autofill can now dig through your Gmail to fill out forms
Google’s Chrome browser is getting an update that will improve its autofill capabilities using AI. On Wednesday, the company said that Chrome will get an advanced version of autofill that will help users fill out online forms with suggestions pulled from their Gmail. That means Chrome could suggest things you might need to know, like a flight confirmation number, membership ID, tracking information, order number, an address, phone number, or credit card number by referencing your email.