<h2>Must know</h2>
<ul>
<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 actively implement post-quantum security measures. They’ve released initial cryptographic standards and recommend hybrid methods as interim solutions, emphasizing immediate action to protect against ‘harvest now, decrypt later’ attacks, as reported by various news outlets. This transition is framed as urgent but long-term, requiring comprehensive security efforts. <a href="https://app.daily.dev/posts/lRZsyJgWP" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Critical ServiceNow AI agent vulnerability disclosed</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, according to AppOmni. ServiceNow has confirmed the vulnerability and released an immediate patch, though no exploitation in the wild has been reported. <a href="https://app.daily.dev/posts/wzAqd15ET" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>CrowdStrike acquires Seraphic Security for browser threat 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, as detailed in official announcements. This integration into CrowdStrike’s Falcon platform aims to correlate endpoint and browser activity, addressing risks from session hijacking and exfiltration via generative AI prompts. <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>MemOS launches open-source memory OS for AI agents</strong>: MemOS, an open-source Memory Operating System for LLMs and AI agents, provides unified long-term memory management with a graph-structured store, multi-modal data support, and claims significant accuracy improvements and token savings over OpenAI Memory. <a href="https://app.daily.dev/feed-by-ids?id=bjqQdB8fJ&amp;id=0rVfXzPjQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Dragonfly CNCF graduates, boosting AI model distribution</strong>: CNCF has graduated Dragonfly, a cloud-native peer-to-peer platform for distributing container images and AI models, which is now widely adopted by major organizations and claims 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>Google’s Gemini 3 Flash accelerates AI workflows</strong>: Gemini 3 Flash is a new AI model optimized for high-volume tasks, offering 3x faster performance than Gemini 2.5 Pro at a lower cost, available across Google’s developer platforms for rapid prototyping and automation. <a href="https://app.daily.dev/posts/LKHPrKtZP" 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>Livewire 4 simplifies component development</strong>: Livewire 4 introduces single-file components, an ‘islands’ feature for isolated rendering, built-in drag-and-drop, and smooth transitions via the View Transitions API, while maintaining backward compatibility. <a href="https://app.daily.dev/feed-by-ids?id=OxhgZzaoQ&amp;id=TQr7IJii0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>BeaglePlay board gets open-source PowerVR graphics</strong>: The BeaglePlay board now supports fully open-source, upstream PowerVR graphics drivers, addressing long-standing community demands for transparency and compatibility from Imagination Technologies. <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 centralized Apache Spark-based system called FINUDP, aiming to eliminate metric inconsistencies with multi-layered quality control and machine learning 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>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI solves math problems, but humans still needed to, you know, think</strong>: OpenAI’s GPT 5.2 is apparently a math whiz, solving 11 of 15 newly resolved Erdős problems. Terence Tao, a human mathematician, graciously admits AI excels at ‘obscure, straightforward’ problems, but humans still need to verify proofs. So, AI can do the grunt work, but don’t expect it to invent new math; that’s still our job, for now. <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>Anthropic’s Cowork: great for enterprise, terrible for security</strong>: Anthropic released Cowork, a feature for Claude Code to automate enterprise tasks, but conveniently highlights its susceptibility to prompt injection attacks and destructive commands. They claim user confirmation mitigates risks, but then their other product, Claude Cowork, was exposed to file exfiltration via indirect prompt injection without user approval. Anthropic’s response? Users should monitor for 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>Quantum computing claims challenged by… a dog?</strong>: Researchers supposedly replicated IBM quantum factorization results using a dog, suggesting overhyped IBM achievements. While this sounds like a satirical jab at quantum progress, it highlights the ongoing skepticism and the need for rigorous, independently verifiable results in a field prone to sensationalism. Maybe the dog just had a really good algorithm. <a href="https://app.daily.dev/feed-by-ids?id=7Bd1YG4JM&amp;id=FBqZ5fVz5" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
