<h2>Must know</h2>
<ul>
<li><strong>Google shares journalist data with ICE</strong>: Google provided ICE with extensive personal data of a student journalist, including bank and credit card numbers, without prior notice. This was in response to an administrative subpoena, which doesn’t require judicial approval. The Electronic Frontier Foundation (EFF) and ACLU are calling on tech companies to resist such requests and notify users, highlighting a concerning trend of increased government data demands. <a href="https://app.daily.dev/posts/nszZi8Hk4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>LLM safety guardrails vulnerable to training prompt exploitation</strong>: Microsoft researchers found a technique, GRP-Obliteration, that bypasses safety guardrails in 15 different LLMs by exploiting reinforcement learning methods. A single, seemingly mild training prompt was enough to make models permissive across harmful categories. This is a significant concern for enterprises fine-tuning open-weight models, emphasizing the need for continuous safety evaluation rather than assuming alignment is static. <a href="https://app.daily.dev/posts/6Gbu8zodz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Docker hardened images now free</strong>: Docker has made its Hardened Images (DHI) free, covering over 1,000 images including Alpine, Debian, databases, and runtimes. This aims to establish a security “waterline” where Docker manages vulnerabilities below, and users handle application-layer issues. DHI includes VEX attestations for automated CVE suppression and supports compliance frameworks like ISO 27001 and SOC 2, integrating with existing scanners to reduce vulnerability noise. <a href="https://app.daily.dev/posts/qrz4zhyW1" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Go 1.26 released with performance and security enhancements</strong>: Go 1.26 was released with the Green Tea garbage collector by default, reducing cgo overhead by 30%, adding experimental SIMD support, and introducing a <code>runtime/secret</code> package for secure memory handling. <a href="https://app.daily.dev/posts/5IorXmNcp" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Node.js updates current and LTS releases</strong>: Node.js released versions 25.6.1 (Current) and 24.13.1 (LTS), with the LTS version adding Python 3.14 build support, promoting heap snapshot flags to stable, and including security patches for TLSSocket error handling. <a href="https://app.daily.dev/posts/CeFkKNaVQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>LinkedIn revamps service discovery for lower latency</strong>: LinkedIn replaced its Zookeeper-based service discovery with a new Kafka and gRPC/xDS architecture, achieving 10x better P50 latency and 6x better P99 latency for hundreds of thousands of service instances. <a href="https://app.daily.dev/posts/WAH3amRwM" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Amazon Redshift enhances automatic optimizations</strong>: Amazon Redshift now allows allocating extra compute resources for automatic optimization features like Automatic Table Optimization, Auto Vacuum, and Auto Analyze, ensuring they run without impacting user workloads. <a href="https://app.daily.dev/posts/7o3DKKn6E" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Mistral Voxtral Realtime 4B C implementation released</strong>: A pure C implementation of Mistral’s Voxtral Realtime 4B speech-to-text model was released, featuring zero external dependencies, Metal GPU acceleration for Apple Silicon, and streaming output with incremental audio processing. <a href="https://app.daily.dev/posts/oP3oVz2Y3" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>CrossOver 26 brings Wine 11.0 to Linux and macOS</strong>: CrossOver 26, built on Wine 11.0, was released to enable Windows applications and games on Linux and macOS, including NTSync support for improved game performance on Linux and enhanced Wayland compatibility. <a href="https://app.daily.dev/posts/Rhd5Rku0N" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Jakarta EE 12 M2 introduces unified query language</strong>: Jakarta EE 12 Milestone 2 introduces Jakarta Query, a unified query language for SQL and NoSQL databases, consolidating JPQL and JDQL, and establishing Java 21 as its baseline. <a href="https://app.daily.dev/posts/y8GIbPRTH" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI singularity predicted for 2034, or maybe not</strong>: A mathematical analysis predicts a “singularity date” of July 2034 based on AI progress metrics, but notes only arXiv papers on AI emergence show genuine hyperbolic curvature, while capability metrics remain linear. So, while the AI hype train is full steam ahead, the actual technical progress might be more of a gentle slope than a rocket launch. But hey, a “social singularity” is happening now, apparently, with mass layoffs and market concentration, which sounds a lot like business as usual, just with more AI buzzwords. <a href="https://app.daily.dev/posts/86v6ixcll" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI tools lead to burnout, not just productivity</strong>: A Harvard Business Review study found AI tools led to burnout and expanded to-do lists for tech workers, with work bleeding into personal hours, despite no management pressure. Apparently, if you give people tools that make them <em>feel</em> like they can do more, they’ll just do more, even if the actual productivity gains are modest. Who knew that making work easier could just mean more work? Revolutionary. <a href="https://app.daily.dev/posts/RvA4dXVbd" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents in messaging apps can spill secrets</strong>: AI agents integrated into messaging apps are vulnerable to zero-click data exfiltration attacks, where prompt injection can trick them into generating malicious URLs that leak sensitive data when link previews are fetched. So, your super-smart AI assistant in Slack can be easily conned into spilling your company’s secrets just by showing a link. Because, of course, the solution to all our security problems is to add more layers of complexity that can be trivially exploited. <a href="https://app.daily.dev/posts/FJDo46xiv" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
