<h2>Must know</h2>
<ul>
<li><strong>Critical ServiceNow vulnerability bypassed MFA and SSO</strong>: AppOmni discovered ‘BodySnatcher’ (CVE-2025-12420), a critical ServiceNow flaw allowing unauthenticated attackers to impersonate users with just an email, bypassing MFA and SSO. This could lead to rogue AI agents with escalated privileges accessing external environments. ServiceNow has patched the vulnerability, but it 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 FBI, NIST, and CISA, are moving from awareness to active implementation of post-quantum security measures. NIST has released initial standards, recommending hybrid cryptographic methods as interim solutions. This multi-year process is framed as urgent, requiring immediate action to protect against ‘harvest now, decrypt later’ attacks, as reported by various news outlets. <a href="https://app.daily.dev/posts/lRZsyJgWP" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>CrowdStrike acquires Seraphic Security for browser protection</strong>: CrowdStrike is acquiring Seraphic Security for approximately $400 million, marking its fourth acquisition since August. Seraphic specializes in enterprise browser security, isolating JavaScript engines and randomizing memory addresses to prevent exploitation. This integration into CrowdStrike’s Falcon platform aims to correlate endpoint and browser activity, addressing risks like session hijacking and data exfiltration through generative AI prompts, according to official announcements. <a href="https://app.daily.dev/posts/AKJAUyPZD" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>AI agents improve code review and context discovery</strong>: Cursor now uses dynamic context discovery to reduce LLM token usage by 46.9% for AI agents, while GitHub Security Lab’s Taskflow Agent offers an open-source framework for AI-powered security research using natural language and MCP interfaces. <a href="https://app.daily.dev/feed-by-ids?id=nGGCLLBtk&amp;id=zVkqeCNSc" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Open-source PowerVR graphics drivers for BeaglePlay</strong>: The BeaglePlay board now supports fully open-source, upstream PowerVR graphics drivers, 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><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 and process millions of booking transactions daily with hourly updates. <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>CNCF graduates Dragonfly for cloud-native distribution</strong>: Dragonfly, a peer-to-peer image and file distribution platform for container images and AI models, has graduated from CNCF, demonstrating production readiness and widespread adoption by major organizations. <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>Firefox 147 enhances privacy and performance</strong>: Firefox 147 introduces automatic Picture-in-Picture, reorganized tab settings, enhanced WebGPU support for Apple Silicon, 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 adds support for new graphics hardware</strong>: Intel’s Compute Runtime version 26.01.36711.4 adds initial support for Crescent Island and Nova Lake S graphics hardware, maintaining Level Zero and OpenCL support for new and past Intel GPUs. <a href="https://app.daily.dev/posts/54Uqeadiv" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Livewire 4 introduces single-file components and islands</strong>: Livewire 4 is out with single-file components, isolated rendering ‘islands’ for performance, built-in drag-and-drop, and smooth transitions via the View Transitions API, while maintaining backward compatibility. <a href="https://app.daily.dev/posts/OxhgZzaoQ" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI agents: the new ‘senior developer’ skill, apparently</strong>: According to some, senior developers are now ‘vibe coding’ with AI tools because they’re so good at delegation and orchestration, not because they can’t code. Apparently, this means AI agents are forcing everyone to develop management skills earlier. So, if you’re not orchestrating your AI, you’re just a junior, even if you’re actually writing code that works, unlike some of the AI-generated garbage out there. <a href="https://app.daily.dev/feed-by-ids?id=WNDZlg59o&amp;id=IcLyTC7xk&amp;id=O3swuQWEt" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic’s Cowork: great for enterprise, not so much for security</strong>: Anthropic released Cowork, a new feature for Claude Code, to automate enterprise tasks, but it’s already susceptible to prompt injection attacks and destructive commands. While they claim user confirmation mitigates risks, they also blocked API access for third-party apps, forcing users onto their platform. So, they’re making it easier for enterprises to automate tasks, but harder for developers to secure them, and then blaming users for not monitoring ‘suspicious activity.’ Classic. <a href="https://app.daily.dev/feed-by-ids?id=GMZHoSF60&amp;id=QvZHdydfi&amp;id=U2HzInRUP&amp;id=ejPwSyHLL&amp;id=G7pZqTe0S&amp;id=jaKSkIh92&amp;id=rAHVvRPEj" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI solves math problems, but humans still need to check its homework</strong>: OpenAI’s GPT 5.2 is apparently solving Erdős problems and LLMs are achieving 80-90% accuracy on statistical coding tasks. But wait, humans still need to verify proofs and review code because AI makes errors, just like humans. So, AI is a ‘force multiplier’ for well-architected systems, not a replacement for senior engineers. In other words, it’s great for the easy stuff, but don’t trust it with anything important. <a href="https://app.daily.dev/feed-by-ids?id=8o6H9wDXV&amp;id=mhyD4ij4A" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
