<h2>Must know</h2>
<ul>
<li><strong>Critical ServiceNow AI agent vulnerability disclosed</strong>: AppOmni discovered ‘BodySnatcher’ (CVE-2025-12420), a critical flaw in ServiceNow’s AI agent integration that allowed unauthenticated attackers to impersonate users with just an email address, bypassing MFA and SSO. This could lead to rogue AI agents with escalated privileges accessing external environments. ServiceNow has since released a patch, but the incident highlights the growing attack surface with enterprise AI adoption. <a href="https://app.daily.dev/posts/wzAqd15ET" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>US agencies push post-quantum security transition</strong>: U.S. federal agencies, including the FBI, NIST, and CISA, are moving beyond awareness to active implementation of 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 is framed as urgent, requiring immediate action for cryptographic inventory and integration, as reported by government sources. <a href="https://app.daily.dev/posts/lRZsyJgWP" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic’s Claude Code API changes spark developer backlash</strong>: Anthropic has restricted API access for Claude Code from all subscription plans, limiting its use to Anthropic’s own tools and effectively blocking third-party applications like Cursor and Open Code. This move, confirmed by developer complaints on GitHub, tightly couples workflows to Anthropic’s platforms, reducing flexibility for users. Meanwhile, Anthropic also released Cowork, a new feature for Claude Code to automate enterprise tasks, emphasizing its capabilities while acknowledging prompt injection risks. <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><strong>Dragonfly graduates from CNCF, boosts AI model distribution</strong>: Dragonfly, a cloud-native peer-to-peer image and file distribution platform, has graduated from CNCF, demonstrating production readiness and widespread adoption by major organizations for container images and AI models, claiming up to 90% storage bandwidth savings. <a href="https://app.daily.dev/feed-by-ids?id=0rVfXzPjQ&amp;id=bjqQdB8fJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Agoda centralizes financial data with Apache Spark</strong>: Agoda consolidated its financial data pipelines into a unified Apache Spark-based platform, FINUDP, to eliminate metric inconsistencies, processing millions of booking transactions daily with machine learning for anomaly detection. <a href="https://app.daily.dev/feed-by-ids?id=SIYRILab7&amp;id=WuRILab7" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>BeaglePlay board gets fully open-source PowerVR graphics</strong>: The BeaglePlay board now supports fully open-source, upstream PowerVR graphics drivers, a significant shift that addresses long-standing demands from the Linux and open-source communities for transparency and improved 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><strong>Livewire 4 introduces single-file components and ‘islands’</strong>: Livewire 4 is out, bringing single-file components for scoped CSS/JS, an ‘islands’ feature for isolated rendering and performance boosts, and built-in drag-and-drop functionality, all while maintaining backward compatibility. <a href="https://app.daily.dev/posts/OxhgZzaoQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Firefox 147 enhances privacy and platform support</strong>: Firefox 147 launched with automatic Picture-in-Picture, reorganized tab settings, enhanced WebGPU support for Apple Silicon Macs, zero-copy hardware video playback for Windows AMD users, and stricter Enhanced Tracking Protection. <a href="https://app.daily.dev/posts/f7pzHnDhN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Intel Compute Runtime adds Crescent Island and Nova Lake S support</strong>: Intel’s Compute Runtime version 26.01.36711.4 introduces initial support for its upcoming Crescent Island and Nova Lake S graphics hardware, maintaining Level Zero and OpenCL support for new and older Intel GPUs. <a href="https://app.daily.dev/posts/54Uqeadiv" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>React ecosystem sees updates and new tools</strong>: React frameworks are converging in features, with Next.js 16.1 improving filesystem cache, Immer 11.1 boosting performance, and new tools like Waku 1.0 alpha and Voltra 1.0 for React Native Live Activities emerging. <a href="https://app.daily.dev/feed-by-ids?id=Ii8I7s3X0&amp;id=oWHYd6BqQ" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI solves ‘Erdős problems’ while human errors persist</strong>: OpenAI’s GPT 5.2 is apparently solving obscure mathematical problems, with claims of 11 new solutions since Christmas. Meanwhile, actual human researchers are still making mathematical errors in 62% of population modeling studies. So, AI can tackle the ‘long tail’ of math, but don’t expect it to replace human judgment or fix our own sloppy work anytime soon, as Terence Tao wisely points out. <a href="https://app.daily.dev/feed-by-ids?id=8o6H9wDXV&amp;id=mhyD4ij4A" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Quantum computing claims challenged by a dog, apparently</strong>: Researchers are ‘critiquing’ IBM’s quantum computing claims by replicating factorization results with a dog. Yes, a dog. This, alongside an undergraduate outperforming a Turing Award winner in hash tables, and Valve’s CPU scheduler being adopted by Meta, makes you wonder if we’re making actual progress or just generating more clickbait. The only thing truly reliable here is Rice University’s actual science on molecular energy transfer. <a href="https://app.daily.dev/feed-by-ids?id=7Bd1YG4JM&amp;id=FBqZ5fVz5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agent vulnerabilities: a feature, not a bug?</strong>: Microsoft Copilot had a vulnerability allowing data exfiltration via a single malicious URL click, which they’ve patched. But Anthropic’s Claude Cowork is still vulnerable to indirect prompt injection, letting attackers steal user data. Anthropic’s response? ‘Users should monitor for suspicious activity.’ So, they market an AI agent to non-technical users, then tell them to become security experts. Classic. <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>
</ul>
