<h2>Must know</h2>
<ul>
<li><strong>Ladybird browser pivots to Rust with AI assistance</strong>: Ladybird, an independent web browser project, is ditching Swift for Rust, primarily for memory safety. Lead developer Andreas Kling used AI agents like Claude Code and Codex to translate 25,000 lines of C++ to Rust in just two weeks, a task estimated to take months manually. The AI-assisted process, described as a human-directed workflow with adversarial review, resulted in byte-for-byte identical output for key components like the JavaScript engine’s lexer and parser, with zero test regressions. <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. The attackers, described as unsophisticated, leveraged generative AI tools for reconnaissance, exploitation instructions, and scripting. They exploited exposed management interfaces, weak credentials, and lack of MFA, moving laterally to Active Directory and Veeam servers, aiming for ransomware deployment. AWS emphasizes that basic security hygiene would have prevented most of these attacks. <a href="https://app.daily.dev/posts/azBzuZltn" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>ZeroDayRAT mobile spyware emerges, bypassing MFA</strong>: ZeroDayRAT, a new mobile spyware family sold openly on Telegram for $2,000, targets Android 5-16 and iOS up to 26. It offers capabilities like keylogging, SMS control (enabling MFA bypass), microphone access, and credential theft. Security researchers warn it commoditizes nation-state-level capabilities, posing significant risks to enterprises with BYOD policies and individuals, as it requires no technical expertise to use. <a href="https://app.daily.dev/feed-by-ids?id=Q0yMWpjwU&amp;id=9TGsm3aiu" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<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. <a href="https://app.daily.dev/posts/NOSp6Xv2e" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kubernetes 1.35 strengthens AI/ML workload support</strong>: Kubernetes v1.35 (‘Timbernetes’) introduces features like stable in-place Pod resource resize and gang scheduling support, strengthening its role as an operating system for AI/ML workloads. <a href="https://app.daily.dev/feed-by-ids?id=mAyDHBOpw&amp;id=nytQTToPj" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI releases gpt-realtime-1.5 for real-time voice AI</strong>: OpenAI released gpt-realtime-1.5 via its Realtime API, focusing on strengthening voice workflows with improved instruction following, tool calling, and multilingual accuracy for low-latency speech-to-speech interactions. <a href="https://app.daily.dev/posts/6I1BavGOr" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitHub introduces Agent HQ for unified AI coding agents</strong>: GitHub introduced Agent HQ, a unified platform for coding agents that integrates Claude, Codex, and GitHub Copilot, allowing users to run and compare multiple frontier agents simultaneously with a single subscription. <a href="https://app.daily.dev/posts/CxN7d4Xat" 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 designed to support over 70 languages, including low-resource ones, and capable of running on mobile devices. <a href="https://app.daily.dev/posts/QusoaclzE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google’s Firefly protocol achieves sub-10ns clock sync</strong>: Google’s Firefly is a software-driven clock synchronization protocol achieving sub-10ns NIC-to-NIC accuracy in data centers using commodity hardware, crucial for high-frequency trading and distributed databases. <a href="https://app.daily.dev/posts/q2urNBdgF" 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 browser environment for AI agents to interact with the web, allowing them to safely execute bash, Python, and JavaScript commands. <a href="https://app.daily.dev/posts/AY0JUdVv6" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Anthropic cries foul over “industrial-scale distillation attacks”</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 capabilities. Anthropic frames this as a national security threat, conveniently aligning with ongoing U.S. debates on AI chip export controls. Funny how these “national security threats” always seem to benefit the accuser’s market position, especially when they themselves are facing copyright infringement lawsuits for training on scraped content. <a href="https://app.daily.dev/feed-by-ids?id=YcZmCF704&amp;id=s76VxSBCH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google DeepMind “solves” image generation problem, details TBD</strong>: Google DeepMind has “reportedly solved a significant problem in image generation,” with the fix described as “embarrassingly elegant” by a Twitter user. No actual details, no official announcement, just a retweet with maximum hype. It’s almost as if they want us to believe in magic without showing the trick. I’m sure this “embarrassingly elegant” solution will be revealed any day now, right after they figure out how to make it sound even more revolutionary. <a href="https://app.daily.dev/posts/fzJvtM3B4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI updates AI coding evaluation standard, old one was “saturated”</strong>: OpenAI announced a change in its recommended standard for frontier coding evaluations, now pushing SWE-bench Pro because the previous benchmark, SWE-bench Verified, became “saturated due to test-design issues and contamination from public repositories.” So, after hyping up a benchmark, they’re now telling us it was flawed all along, and here’s a new one that’s definitely, absolutely, truly rigorous this time. It’s almost like they’re moving the goalposts as soon as their models start looking too good on the old ones. <a href="https://app.daily.dev/posts/oZGyDrjcX" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
