<h2>Must know</h2>
<ul>
<li><strong>AI coding agents are getting more capable and more dangerous</strong>: A shift from simple AI assistance to autonomous ‘agent swarms’ is creating new security risks, including a recent incident where an unsupervised agent leaked secrets to a public registry. Experts at QCon London highlighted that while tools like Claude Code can now build entire working apps and run their own tests, they also significantly increase development costs and introduce vulnerabilities like prompt injection. To manage this, organizations are increasingly using AI gateways and abstraction layers to avoid being trapped by the rapid cycle of model updates from OpenAI and Google. <a href="https://app.daily.dev/feed-by-ids?id=Ur5RHzsG2&amp;id=VPYYe9XNN&amp;id=bCQotnt4R&amp;id=eNry4HYqW&amp;id=f6EHrsEI8&amp;id=sow9Hx2sg&amp;id=xNA0t95XE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New laws force a privacy showdown for mobile operating systems</strong>: GrapheneOS has publicly refused to comply with new age verification laws in Brazil and California, stating it will never collect personal identification even if it means being banned from those regions. These laws require operating systems to track user age and share that data with apps, a move critics argue creates a permanent surveillance infrastructure. This stance complicates GrapheneOS’s upcoming hardware partnership with Motorola, as the OS prioritizes technical privacy over regional legal compliance. <a href="https://app.daily.dev/feed-by-ids?id=3oUfmyp1R&amp;id=f2c1Srdfh&amp;id=u4yVQPhrL" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI faces massive copyright lawsuit from major publishers</strong>: Encyclopedia Britannica and Merriam-Webster have sued OpenAI, alleging that nearly 100,000 of their articles were used to train AI models without permission. The lawsuit claims ChatGPT often reproduces their content verbatim and generates ‘hallucinations’ that are falsely attributed to the publishers, damaging their reputation. This follows a trend of high-profile legal challenges from the New York Times and others that could fundamentally change how AI companies source their training data. <a href="https://app.daily.dev/posts/03LaAXXJF" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Vercel Workflow now features automatic end to end encryption for all user data</strong>: The update secures inputs, return values, and payloads using unique keys per deployment to prevent unauthorized data access. <a href="https://app.daily.dev/posts/0r6jV9CyY" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>A new inference engine runs massive AI models on standard MacBooks</strong>: Flash-MoE allows a 397-billion parameter model to run at usable speeds on consumer hardware by streaming data directly from the SSD. <a href="https://app.daily.dev/posts/xsGaEEFUf" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux 7.0 is approaching a stable release this April</strong>: Linus Torvalds released the fifth candidate for version 7.0, noting the development cycle is calming down despite some driver workarounds. <a href="https://app.daily.dev/posts/1tWnyX7nj" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The popular MkDocs project is fracturing due to maintainer infighting</strong>: The documentation tool ecosystem has split into three competing versions after the original founder returned to start a private, incompatible rewrite. <a href="https://app.daily.dev/posts/GedRmjUpE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>BART train outages are being caused by failing 1970s era networking gear</strong>: Recent service halts for 175,000 riders were traced to aging components that, when they fail, make trains invisible to operators. <a href="https://app.daily.dev/posts/3egO5D35k" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Palantir secures a controversial UK financial fraud detection contract</strong>: The deal allows Palantir to analyze sensitive phone recordings and emails for the FCA, sparking significant privacy concerns among insiders. <a href="https://app.daily.dev/posts/zso4fAO8p" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>A developer successfully stored persistent data inside a computer mouse</strong>: By reverse-engineering the Logitech protocol, a researcher found the DPI register can be used as a tiny, cross-computer storage device. <a href="https://app.daily.dev/posts/uG0HL5oPv" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The zero standby power charger that actually uses power</strong>: Belkin is marketing a ‘zero standby’ USB-C charger that actually draws 3 milliwatts, which jumps to 30 milliwatts the second you plug in a high-end cable with nothing attached to it. To make matters worse, the ‘smart’ logic meant to save power is reportedly causing the charger to randomly cut out while people are actually trying to use it. Maybe we should just use a physical power switch instead of over-engineering a solution that fails at its one job. <a href="https://app.daily.dev/posts/vdXu1e0OL" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cursor tries to hide its Chinese AI homework</strong>: The popular AI code editor Cursor launched ‘Composer 2’ with a lot of hype about its ‘frontier-level’ intelligence, conveniently forgetting to mention it’s actually built on an open-source model from China’s Moonshot AI. After an X user caught them red-handed, the VP of education had to admit the ‘mistake’ in disclosure. It turns out ‘frontier intelligence’ is often just someone else’s model with a fresh coat of marketing paint. <a href="https://app.daily.dev/posts/2O2HmWwMM" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The death of coding is the new fast fashion</strong>: Pundits are currently obsessed with the idea that AI is ‘killing the craft’ of coding, turning software into disposable ‘vibe-coded’ garbage that we’ll just throw away. This narrative ignores the fact that English is a terrible language for precise specifications, and ‘vibe coding’ usually just results in expensive spaghetti code that no one can fix. AI isn’t killing coding any more than the calculator killed mathematics; it’s just making the boring parts faster so we can find new ways to break things. <a href="https://app.daily.dev/feed-by-ids?id=AKKujKavD&amp;id=3336MSPgf" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
