<h2>Must know</h2>
<ul>
<li><strong>Google shares journalist’s data with ICE without notice</strong>: Google provided extensive personal data, including bank details and IP addresses, of student journalist Amandla Thomas-Johnson to ICE in response to an administrative subpoena. Crucially, Google did not notify Thomas-Johnson, denying her the chance to challenge the request. This highlights a significant privacy concern, as administrative subpoenas bypass judicial approval and tech companies are not legally obligated to comply or notify users, according to the Electronic Frontier Foundation (EFF) and ACLU. <a href="https://app.daily.dev/posts/nszZi8Hk4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI safety vulnerability allows disabling LLM guardrails</strong>: Microsoft researchers discovered ‘GRP-Obliteration,’ a technique that uses a single benign training prompt to systematically disable safety guardrails in 15 major language models. This method, which exploits Group Relative Policy Optimization, can increase a model’s attack success rate from 13% to 93% across 44 harmful categories. The vulnerability primarily affects open-weight models during fine-tuning, urging enterprises to treat model customization as a controlled risk and continuously evaluate safety. <a href="https://app.daily.dev/posts/6Gbu8zodz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>BeyondTrust RCE flaw patched</strong>: BeyondTrust released patches for a critical 9.9-severity Remote Code Execution (RCE) flaw (CVE-2026-1731) affecting its self-hosted Remote Support and Privileged Remote Access products. The vulnerability requires no authentication for exploitation, posing a significant risk for system compromise and data exfiltration. Hacktron AI discovered the flaw using AI-enabled variant analysis, underscoring the importance of immediate patching for the estimated 8,500 affected on-premises instances. <a href="https://app.daily.dev/posts/5QKIgzQLm" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Pure C implementation of Mistral Voxtral 4B released</strong>: Antirez released a pure C implementation of Mistral’s Voxtral Realtime 4B speech-to-text model, featuring zero external dependencies, Metal GPU acceleration for Apple Silicon, and streaming output with a rolling KV cache for unlimited audio processing. <a href="https://app.daily.dev/posts/oP3oVz2Y3" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents transform development and operations, raise security concerns</strong>: AI agents are moving from assistive tools to autonomous systems, impacting software development and operations, but also introducing new security risks like unauthorized data access and prompt injection, with 59% of IT professionals anticipating AI-driven cyber threats. <a href="https://app.daily.dev/feed-by-ids?id=0V8Ax2TZk&amp;id=1Om5VKtbo&amp;id=2BkIn6Msl&amp;id=2aKUMsJl0&amp;id=2aVoZAUrh&amp;id=41LADl1IG&amp;id=5TbloOAVZ&amp;id=7UDJV3reE&amp;id=9LBPutt5i&amp;id=A0wRC8dnR" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux 7.0 kernel improves io_uring and scheduler</strong>: Linux 7.0 introduces BPF filtering for io_uring, resolving a long-standing incompatibility with seccomp in containers, and merges significant scheduler improvements for high core count systems with streamlined preemption modes. <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>FastAPI Cloud simplifies deployment with single command</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/feed-by-ids?id=37dupllnW&amp;id=5Mzv43k5F&amp;id=XBUb8C6g8&amp;id=sRoqluNnA" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>CrossOver 26 released with Wine 11.0</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 macOS game compatibility. <a href="https://app.daily.dev/posts/Rhd5Rku0N" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Solidity, Ruby, and Spring ecosystems update</strong>: The Solidity team launched its annual developer survey to shape the 2026 roadmap, JRuby 10.0.3.0 was released, and ‘This Week in Spring’ provided a roundup of Spring Framework news, including new Ruby gems for AI agent frameworks. <a href="https://app.daily.dev/feed-by-ids?id=8mu4nh6xg&amp;id=AvOOFdAE3&amp;id=CJ8030TJe&amp;id=R3XBh5Yuj" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Hermes V1 now default in React Native 0.84</strong>: React Native 0.84 makes Hermes V1 the default JavaScript engine, delivering automatic performance improvements, reduced memory usage, and enabling precompiled iOS binaries by default to speed up build times. <a href="https://app.daily.dev/posts/oBV7ttAiR" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI facial analysis predicts salary and career trajectory</strong>: Academic researchers used AI facial analysis on LinkedIn photos of over 96,000 MBA graduates to predict salary and career trajectory. While they claim it’s for ‘academic evaluation,’ they also admit HR departments are already adopting such tools. So, we’re just ‘academically evaluating’ how to automate discrimination and ML-launder junk science into hiring practices. Brilliant. <a href="https://app.daily.dev/posts/2nGSglCla" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI will perpetually produce bugs due to math</strong>: An article claims LLMs will ‘perpetually produce bugs’ due to their mathematical foundations, operating through approximation and interpolation, making human oversight essential. Well, no kidding. It’s almost like these things aren’t magic or sentient, and are just really complex statistical models. Who would’ve thought that approximating reality might lead to ‘bugs’? <a href="https://app.daily.dev/posts/71H3ju1zr" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Personal productivity workflow with Gemini AI and Google Sheets</strong>: An unnamed author claims to have simplified their life organization and productivity management using Google Sheets with Gemini AI, boasting a ‘two-minute setup.’ Because apparently, the secret to peak productivity isn’t discipline or focus, but a spreadsheet and an AI that probably just summarizes your to-do list. And ‘two minutes’? That’s barely enough time to open Google Sheets. <a href="https://app.daily.dev/feed-by-ids?id=8E4YEJZ09&amp;id=Nb8ar1wlf" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
