<h2>Must know</h2>
<ul>
<li><strong>Ladybird browser pivots to rust with AI assistance</strong>: The Ladybird browser project is ditching Swift for Rust, primarily for memory safety, after losing a year on Swift. Lead developer Andreas Kling used Claude Code and Codex AI agents to translate 25,000 lines of C++ to Rust in just two weeks, a task estimated to take months manually. This human-directed AI workflow resulted in byte-for-byte identical output and zero test regressions, showing AI’s potential for accelerating large-scale refactoring, as reported by multiple sources. <a href="https://app.daily.dev/feed-by-ids?id=bavH6r0ZC&amp;id=1QrlHS5ue" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-augmented FortiGate firewall compromises reported by AWS</strong>: AWS detected a financially motivated, Russian-speaking threat actor compromising over 600 FortiGate firewall instances across 55+ countries. This group, described as technically unsophisticated, used generative AI tools for reconnaissance, exploitation instructions, and script generation. The attacks leveraged exposed management interfaces, weak credentials, and lack of MFA, highlighting that basic security hygiene would have prevented most of these AI-assisted breaches. <a href="https://app.daily.dev/posts/azBzuZltn" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic accuses chinese AI labs of industrial-scale model distillation</strong>: Anthropic publicly accused Chinese AI labs DeepSeek, Moonshot AI, and MiniMax of ‘industrial-scale distillation attacks’ on its Claude models, claiming they created over 24,000 fraudulent accounts to extract Claude’s capabilities. Anthropic frames this as a national security threat, arguing distilled models may lack safety guardrails and could be misused. This accusation comes amidst ongoing US debates on AI chip export controls to China. <a href="https://app.daily.dev/feed-by-ids?id=YcZmCF704&amp;id=s76VxSBCH" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>GitHub introduces Agent HQ for unified AI coding agents</strong>: GitHub announced Agent HQ, a unified platform for coding agents like Claude, Codex, and GitHub Copilot, allowing users to run and compare multiple frontier agents simultaneously with a single Copilot Pro Plus or Enterprise subscription. <a href="https://app.daily.dev/posts/CxN7d4Xat" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI releases gpt-realtime-1.5 model</strong>: OpenAI released gpt-realtime-1.5 via its Realtime API, focusing on strengthening voice workflows with improved instruction following, enhanced tool calling, and better multilingual accuracy for low-latency, multimodal interactions. <a href="https://app.daily.dev/posts/6I1BavGOr" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Uber open-sources uForwarder kafka consumer proxy</strong>: Uber open-sourced uForwarder, a push-based Kafka consumer proxy designed to handle trillions of messages daily, centralizing offset management and simplifying consumer logic for microservices with features like context-aware routing and adaptive auto-rebalancing. <a href="https://app.daily.dev/posts/NOSp6Xv2e" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>ZeroDayRAT mobile spyware emerges, bypassing MFA</strong>: ZeroDayRAT, a new mobile spyware sold openly on Telegram, targets Android and iOS, offering capabilities like keylogging, SMS control (for MFA bypass), screen recording, and credential theft, making advanced surveillance accessible to unsophisticated attackers. <a href="https://app.daily.dev/posts/Q0yMWpjwU" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Starkiller phishing kit bypasses MFA</strong>: Starkiller, a Phishing-as-a-Service toolkit, bypasses MFA and traditional phishing detection by live-proxying real login pages and capturing authenticated session tokens, lowering the technical barrier for attackers and necessitating behavioral and identity-aware detection. <a href="https://app.daily.dev/posts/9TGsm3aiu" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agent browser sandbox project announced</strong>: A new open-source project, ‘browser sandbox,’ provides a secure, zero-configuration browser environment for AI agents to interact with the web, allowing them to safely execute bash, Python, and JavaScript commands to control the browser. <a href="https://app.daily.dev/posts/AY0JUdVv6" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cohere releases Aya Tiny multilingual models</strong>: Cohere released the Aya Tiny model family, small (~3.3B parameter) multilingual language models supporting over 70 languages, designed to run on mobile devices and serve languages often overlooked by larger models. <a href="https://app.daily.dev/posts/QusoaclzE" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>OpenAI updates AI coding evaluation standard</strong>: OpenAI, in a move that’s surely not self-serving, now recommends SWE-bench Pro for AI coding evaluations, claiming their previous benchmark, SWE-bench Verified, became ‘saturated’ due to ‘test-design issues and contamination.’ So, when your AI gets too good for your own tests, just invent a new one, right? Problem solved, and benchmarks look shiny again! <a href="https://app.daily.dev/posts/oZGyDrjcX" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google DeepMind claims image generation breakthrough</strong>: Google DeepMind supposedly ‘just solved’ a ‘dirty problem’ in image generation with an ‘embarrassingly elegant’ fix, according to a retweet from some random Twitter user. No details, no official announcement, just pure, unadulterated hype. Because if it’s not on Twitter, did it even happen? And if it’s ‘embarrassingly elegant,’ why are they so embarrassed to tell us what it is? <a href="https://app.daily.dev/posts/fzJvtM3B4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic’s Claude Code Security launch and market reaction</strong>: Anthropic launched Claude Code Security, an AI tool for vulnerability detection, causing a ‘notable selloff’ in cybersecurity stocks. Analysts called the market reaction ‘illogical,’ which is a polite way of saying investors panicked over an AI tool that still requires human oversight and hasn’t published its false-positive rates. Apparently, the market believes AI will replace security professionals before Anthropic even tells us how well it actually works. <a href="https://app.daily.dev/posts/bgcFYmVK1" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
