<h2>Must know</h2>
<ul>
<li>Critical security flaws in AI agents exposed: Major vulnerabilities have been found in Microsoft Copilot and Anthropic’s Claude Cowork. Copilot allowed data exfiltration via a single malicious URL click, while Claude Cowork was susceptible to indirect prompt injection, enabling attackers to upload user data to their accounts. Microsoft has patched Copilot, but Claude Cowork remains unpatched, with Anthropic placing responsibility on users, as reported by security researchers. This highlights the ongoing security challenges in AI agent development. <a href="https://app.daily.dev/feed-by-ids?id=G7pZqTe0S&amp;id=jaKSkIh92&amp;id=rAHVvRPEj" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>US agencies push for post-quantum security transition: U.S. federal agencies, including the FBI, NIST, and CISA, are moving beyond awareness to actively implement post-quantum security measures. They’ve released initial cryptographic standards and recommend hybrid methods as interim solutions to counter ‘harvest now, decrypt later’ attacks. This multi-year transition emphasizes immediate action for cryptographic inventory and integration, as detailed in official government guidance. <a href="https://app.daily.dev/posts/lRZsyJgWP" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Anthropic restricts Claude API access for third-party tools: Anthropic has blocked API credentials for all Claude subscription plans from third-party applications, limiting access to its own Claude Code tool. This move, confirmed by developer community backlash and GitHub complaints, significantly reduces flexibility for users of tools like Cursor and Open Code. Meanwhile, Anthropic released Cowork, a new feature for Claude Code, enabling automation of enterprise tasks beyond software development. <a href="https://app.daily.dev/feed-by-ids?id=GMZHoSF60&amp;id=QvZHdydfi&amp;id=U2HzInRUP&amp;id=ejPwSyHLL" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li>New open-source memory OS for AI agents: MemOS, an open-source Memory Operating System for LLMs and AI agents, has launched, offering unified long-term memory management with a graph-structured store, multi-modal data support, and claims of significant accuracy improvements and token savings over OpenAI Memory. <a href="https://app.daily.dev/feed-by-ids?id=0rVfXzPjQ&amp;id=bjqQdB8fJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Dragonfly graduates to full CNCF project: Dragonfly, a cloud-native peer-to-peer image and file distribution platform, has graduated to a full CNCF project, demonstrating widespread adoption and production readiness for scaling container images, OCI artifacts, and AI models. <a href="https://app.daily.dev/feed-by-ids?id=0rVfXzPjQ&amp;id=bjqQdB8fJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>AI tools accelerate learning for non-developers: A UX designer with no prior development experience successfully learned Android and Kotlin using Gemini AI in just 2-3 weeks, building an award-winning smart home app, showcasing AI’s potential to democratize technical skills. <a href="https://app.daily.dev/feed-by-ids?id=P7Lnxz2vQ&amp;id=htOI1bHOh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Firefox 147 enhances privacy and performance: Firefox 147 introduces automatic Picture-in-Picture, enhanced WebGPU support for Apple Silicon Macs, zero-copy hardware video playback for Windows AMD users, and stricter Enhanced Tracking Protection, alongside several security fixes. <a href="https://app.daily.dev/posts/f7pzHnDhN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>BeaglePlay board gets open-source PowerVR graphics: The BeaglePlay board now supports fully open-source, upstream PowerVR graphics drivers, a significant shift addressing long-standing demands from the Linux community for transparency and compatibility. <a href="https://app.daily.dev/feed-by-ids?id=HwIiwOf9Q&amp;id=NjVG9GX8g" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Livewire 4 introduces single-file components and performance boosts: Livewire 4 has been released, featuring single-file components for scoped CSS and JavaScript, an ‘islands’ feature for isolated rendering and performance improvements, and built-in drag-and-drop functionality. <a href="https://app.daily.dev/posts/OxhgZzaoQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Laravel Boost 1.8.10 adds AI prompts for code simplification: Laravel Boost v1.8.10 now includes AI prompts, such as a Code Simplifier based on the Claude plugin and a Livewire v4 upgrade assistant, to help developers refactor and upgrade their Laravel code. <a href="https://app.daily.dev/posts/TQr7IJii0" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li>AI solves ‘Erdős problems’ while humans still verify dog-generated quantum proofs: OpenAI’s GPT 5.2 is apparently solving obscure mathematical problems, while researchers are busy ‘critiquing high-profile IBM quantum computing claims by replicating quantum factorization results using a dog.’ So, AI is either revolutionizing math or we’re still trying to figure out if a canine can do quantum physics. Either way, human oversight is still ‘critical’ because, you know, dogs. <a href="https://app.daily.dev/feed-by-ids?id=7Bd1YG4JM&amp;id=FBqZ5fVz5&amp;id=8o6H9wDXV&amp;id=mhyD4ij4A" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>AI-generated code: functionally correct, architecturally a dumpster fire: A senior Android developer reviewed AI-generated Kotlin code and found it to be ‘functionally correct but architecturally problematic.’ Apparently, AI can write code that ‘works’ but completely misses critical Android-specific concerns like lifecycle management and memory optimization. So, if you want a working app that crashes randomly and eats battery, AI’s got you covered. <a href="https://app.daily.dev/posts/O3swuQWEt" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>The ‘un-teachable’ features for production apps: who knew?: Apparently, production applications need ‘five critical features missing from many tutorials’: analytics, error tracking, uptime monitoring, rate limiting, and automated testing. Who would’ve thought that real-world apps need to, you know, work reliably and not cost a fortune? It’s almost like tutorials focus on the fun stuff and leave the actual engineering to the ‘experts.’ <a href="https://app.daily.dev/feed-by-ids?id=hizCv12EP&amp;id=q1wpmdkFL" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>