<h2>Must know</h2>
<ul>
<li><strong>AI agents introduce critical security vulnerabilities in messaging apps</strong>: AI agents integrated into popular messaging platforms like Slack, Telegram, and Microsoft Teams are vulnerable to zero-click data exfiltration. Researchers at PromptArmor discovered that malicious link previews can trick these agents into revealing sensitive data like API keys and credentials to attacker-controlled servers without any user interaction. This highlights a significant new attack vector for AI systems, especially those with access to internal company data. <a href="https://app.daily.dev/posts/FJDo46xiv" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google shares journalist’s data with ICE without notice</strong>: Google provided extensive personal data, including bank details, IP addresses, and phone numbers, of a student journalist to ICE in response to an administrative subpoena. Crucially, Google did not notify the journalist, Amandla Thomas-Johnson, allowing no opportunity to challenge the request. This incident, highlighted by the Electronic Frontier Foundation (EFF) and ACLU, underscores the ongoing tension between tech companies’ user privacy tools and government agencies’ broad subpoena powers, especially when targeting activists. <a href="https://app.daily.dev/feed-by-ids?id=nszZi8Hk4&amp;id=GZXAyndcW&amp;id=acA7P68UC" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft researchers find new way to disable LLM safety guardrails</strong>: Microsoft researchers have discovered ‘GRP-Obliteration,’ a technique that can disable safety guardrails in 15 major language models. By using a single benign training prompt, this method systematically makes models permissive across all 44 harmful categories, increasing one model’s attack success rate from 13% to 93%. This vulnerability is particularly concerning for enterprises customizing open-weight models, as it requires training access, not just inference-time manipulation, emphasizing the need for continuous safety evaluation. <a href="https://app.daily.dev/posts/6Gbu8zodz" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Pure C implementation of Mistral’s Voxtral 4B speech-to-text model released</strong>: Antirez released a pure C implementation of Mistral’s Voxtral Realtime 4B speech-to-text model with zero external dependencies, Metal GPU acceleration for Apple Silicon, and streaming output, running 2.5 times faster than real-time on an M3 Max chip. <a href="https://app.daily.dev/posts/oP3oVz2Y3" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Docker hardened images now free for over 1,000 images</strong>: Docker has made its Hardened Images (DHI) free, covering over 1,000 images including Alpine, Debian, databases, and runtimes, establishing a security ‘waterline’ and providing VEX attestations for automated CVE suppression. <a href="https://app.daily.dev/posts/qrz4zhyW1" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>FastAPI cloud launches to simplify python web app deployment</strong>: FastAPI Cloud, a new platform backed by Sequoia Capital, simplifies FastAPI application deployment with a single command, handling HTTPS, auto-scaling, and infrastructure complexity on Kubernetes, aiming for a ‘Pythonic’ developer experience. <a href="https://app.daily.dev/posts/37dupllnW" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux 7.0 kernel introduces BPF filtering for io_uring</strong>: Linux 7.0 now supports BPF filtering for io_uring, resolving a long-standing incompatibility with seccomp that previously forced containers to disable io_uring, enabling fine-grained control over operations without sacrificing performance. <a href="https://app.daily.dev/feed-by-ids?id=7pxKDEZxZ&amp;id=RL4bMHXJh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>CrossOver 26 released with wine 11.0 for windows apps on linux and macos</strong>: CodeWeavers released CrossOver 26, a commercial compatibility layer built on Wine 11.0, enabling Windows applications and games to run on Linux and macOS with over 6,000 changes, including NTSync support for Linux and enhanced Wayland compatibility. <a href="https://app.daily.dev/posts/Rhd5Rku0N" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>TutorialKit.rb enables in-browser ruby/wasm tutorials</strong>: Evil Martians introduced TutorialKit.rb, a framework for building interactive Ruby and Rails tutorials that run entirely in a web browser using WebAssembly, providing zero-install learning experiences through enhanced virtual filesystems and custom gem bundling. <a href="https://app.daily.dev/posts/zHyXjBJX2" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>BeyondTrust patches critical RCE flaw</strong>: BeyondTrust released patches for CVE-2026-1731, a critical 9.9-severity Remote Code Execution (RCE) flaw affecting self-hosted Remote Support and Privileged Remote Access products, which requires no authentication for exploitation and was discovered by Hacktron AI. <a href="https://app.daily.dev/posts/5QKIgzQLm" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI doesn’t reduce work, it just intensifies it</strong>: A Berkeley Haas study found that AI tools don’t actually reduce workload; they just make you feel more productive while intensifying work and causing cognitive overload. So, while you’re juggling a dozen AI-generated tasks, remember you’re not working smarter, just harder, and probably burning out faster. Thanks, AI, for the ‘flow state’ that leads straight to exhaustion. <a href="https://app.daily.dev/posts/2nGSglCla" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-powered IPv6 migrations cut hours to seconds, but who cares about cost?</strong>: A university experiment claims generative AI reduced IPv6 migration times from hours to seconds, eliminating errors and cutting cognitive load by 72%. But don’t worry, the CIO is already fretting about AI cost concerns like token pricing and vendor lock-in. Because apparently, saving countless hours and preventing critical errors isn’t worth the potential expense of, you know, actually using the technology. <a href="https://app.daily.dev/posts/e9OZK0ZNf" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>My two-minute google sheets and gemini AI productivity hack</strong>: Someone out there is claiming they’ve revolutionized personal productivity by combining Google Sheets with Gemini AI in a ‘two-minute setup’ that replaces complex tools like Notion. Because, obviously, a spreadsheet and a chatbot are all you need to manage your entire life, and any claim of a ‘two-minute setup’ for anything beyond boiling water is just pure, unadulterated marketing fluff. <a href="https://app.daily.dev/feed-by-ids?id=8E4YEJZ09&amp;id=Nb8ar1wlf" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
