<h2>Must know</h2>
<ul>
<li><strong>Gemini AI model faces extraction attempts and nation-state misuse</strong>: Google has blocked over 100,000 attempts to extract proprietary reasoning capabilities from its Gemini AI model, identifying these as intellectual property theft. These efforts, using model distillation techniques, were coordinated to clone Gemini’s multilingual reasoning. Furthermore, Google’s threat report reveals nation-state actors from China, Iran, North Korea, and Russia are integrating Gemini into attack operations, including social engineering and phishing, with new malware families like HONESTCUE embedding Gemini’s API to generate malicious code that bypasses safety filters. <a href="https://app.daily.dev/posts/OQLlwLpSw" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic secures $30 billion funding amidst AI valuation scrutiny</strong>: Anthropic closed a massive $30 billion Series G funding round, doubling its valuation to $380 billion, driven by a reported $14 billion revenue run rate from enterprise adoption of Claude AI models. This investment, led by GIC and Coatue, signals strong investor confidence in enterprise-grade AI. However, questions linger about the sustainability of such valuations and the actual quality of AI-generated code, with one article questioning Anthropic’s AI-built C compiler’s functionality. <a href="https://app.daily.dev/feed-by-ids?id=0dRrHpMfv&amp;id=2KG0fxG7Z&amp;id=KBiI9wOrx&amp;id=MPMZLgvp3&amp;id=PryOnuBPQ&amp;id=Qmgd9m7GA&amp;id=s5BDVI2HR&amp;id=vELqgGbwK&amp;id=vz4jhHvn8&amp;id=xNryU16Jc" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Odido data breach impacts 6.2 million customers</strong>: Dutch mobile operator Odido disclosed a data breach affecting 6.2 million customers, roughly one-third of the Netherlands’ population. Attackers accessed their customer contact system, compromising names, addresses, phone numbers, dates of birth, bank account numbers, and ID document details. Odido detected and shut down access quickly, reporting the incident to authorities and warning customers about potential phishing and impersonation scams. <a href="https://app.daily.dev/posts/3j2lBB9LH" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Rust-based LLMTrace provides real-time security for AI interactions</strong>: LLMTrace, an open-source transparent proxy built in Rust, offers zero-code observability and security for LLM applications by detecting prompt injection, PII leaks, and cost overruns in real-time. <a href="https://app.daily.dev/posts/mco25QGBm" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cloudflare open-sources rust library for zero-downtime service upgrades</strong>: Cloudflare open-sourced ecdysis, a Rust library enabling zero-downtime service upgrades through graceful process restarts, proven in production for five years across 330+ data centers. <a href="https://app.daily.dev/posts/tdBY3TVs4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Go-powered AI assistant runs on minimal hardware</strong>: PicoClaw is an ultra-lightweight AI assistant written in Go that runs on $10 hardware with less than 10MB RAM, offering chat integrations, web search, and a security sandbox. <a href="https://app.daily.dev/posts/qxYADTM4h" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Orchestra offers self-hosted PaaS for bare metal with Go backend</strong>: Orchestra is an open-source, self-hosted PaaS built with Go and Next.js, automating server provisioning and application deployment for bare metal infrastructure without requiring agents. <a href="https://app.daily.dev/posts/0nUdXf1b9" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Laravel-PDF v2 integrates Cloudflare for dependency-free PDF generation</strong>: Spatie released Laravel-PDF v2, featuring a Cloudflare API driver that allows PDF generation without local dependencies like Node.js or Docker, alongside new features like queued PDF generation and metadata support. <a href="https://app.daily.dev/posts/cK3euMd6K" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>FrankenPHP v1.11.2 boosts performance and fixes critical security flaws</strong>: FrankenPHP v1.11.2 is a critical security release fixing vulnerabilities and upgrading to Go 1.26, delivering 10-40% faster garbage collection and ~30% faster CGO calls. <a href="https://app.daily.dev/posts/T7AGr1PoJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Auth0 updates B2B plans for greater flexibility</strong>: Auth0 updated its B2B plans, now offering one Enterprise Connection in the Free plan for POCs and allowing incremental additions of Enterprise Connections and M2M token add-ons in the B2B Essentials plan for greater flexibility. <a href="https://app.daily.dev/posts/UzL5CQ1eX" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Minimax launches ‘cost-effective’ M2.5 AI model</strong>: Minimax launched its M2.5 AI model, claiming it’s over 90% cheaper than competitors and benchmarked as the “top open model for coding tasks” by OpenHands. Apparently, this 230B parameter model with 10B active parameters achieves competitive performance through “scaled reinforcement learning across hundreds of thousands of training environments.” Because, you know, more environments always mean better code, not just more places for bugs to hide. And who needs independent verification when you have a benchmark from “OpenHands”? <a href="https://app.daily.dev/posts/VQt2ETdNG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kubernetes is ‘overkill’ for most apps, says team running on docker compose</strong>: A team claims to run a production observability platform handling 500k logs/day on a single server using Docker Compose, arguing Kubernetes is “overkill for 99% of apps.” They boast 99.8% uptime at $14/month. Sure, for a single server, Docker Compose is great. But let’s see how that scales when you actually need resilience, multi-region deployments, or more than a handful of services. “Overkill” often just means “I haven’t needed it yet.” <a href="https://app.daily.dev/posts/PNwYwcQ5e" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Spotify developers shipping features from slack with AI</strong>: Spotify developers are reportedly using Claude Code to ship features and fix bugs without manual code, with top developers fixing bugs from phones and shipping 50+ features from Slack since December. And Claude Code supposedly accounts for 4% of GitHub public commits. Right, because the pinnacle of software engineering is fixing bugs from your phone and pushing features from Slack, all while an AI does the actual coding. What could possibly go wrong with that kind of “efficiency”? <a href="https://app.daily.dev/feed-by-ids?id=0dRrHpMfv&amp;id=2KG0fxG7Z&amp;id=KBiI9wOrx&amp;id=MPMZLgvp3&amp;id=PryOnuBPQ&amp;id=Qmgd9m7GA&amp;id=s5BDVI2HR&amp;id=vELqgGbwK&amp;id=vz4jhHvn8&amp;id=xNryU16Jc" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
