<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 allows 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 released a patch, as reported by AppOmni, highlighting 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. NIST has released initial standards, recommending hybrid cryptographic methods as interim solutions. This multi-year process emphasizes immediate action to protect current infrastructure and safeguard quantum research, supply chains, and workforce against ‘harvest now, decrypt later’ threats, according to official government guidance. <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 threat protection</strong>: CrowdStrike is acquiring Seraphic Security for approximately $400 million, marking its fourth acquisition since August. Seraphic specializes in enterprise browser security, using an abstraction layer to isolate JavaScript engines and randomize memory addresses against exploits. CrowdStrike plans to integrate Seraphic into its Falcon platform, combining endpoint data with browser activity for real-time correlation and dynamic access controls, addressing risks like session hijacking and data exfiltration via AI prompts, as detailed in company 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>CNCF graduates Dragonfly for cloud-native image distribution</strong>: Dragonfly, a peer-to-peer image and file distribution platform for container images, OCI artifacts, and AI models, has graduated from CNCF, demonstrating production readiness and widespread adoption by major organizations like Alibaba and Datadog, saving up to 90% storage bandwidth and reducing launch times. <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>AI agent for Kubernetes diagnostics on EKS</strong>: AIOps Sherlock is an open-source diagnostics assistant using multi-agent AI architecture to troubleshoot Kubernetes workloads on Amazon EKS, employing specialized AI agents to analyze pod health, CloudWatch metrics, and DynamoDB performance. <a href="https://app.daily.dev/posts/HuWoOJXv8" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Agoda centralizes financial data with Apache Spark</strong>: Agoda consolidated multiple financial data pipelines into FINUDP, a centralized Apache Spark-based platform, to eliminate metric inconsistencies, processing millions of booking transactions daily with multi-layered quality control and 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>New open-source memory OS for LLMs and AI agents</strong>: MemOS is an open-source Memory Operating System for LLMs and AI agents, providing unified long-term memory management with graph-structured storage, multi-modal data support, and claims of significant accuracy improvement and token savings over existing solutions. <a href="https://app.daily.dev/posts/bjqQdB8fJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>React frameworks converge, AI integration explored</strong>: React framework creators noted a convergence in features, giving developers more choice, with all major frameworks reportedly investigating AI integration using Model Context Protocol (MCP) endpoints, and new tools like Waku 1.0 alpha and Voltra 1.0 for React Native emerging. <a href="https://app.daily.dev/feed-by-ids?id=Ii8I7s3X0&amp;id=oWHYd6BqQ" 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, addressing long-standing demands from the Linux community for transparency and compatibility, a significant shift from Imagination Technologies’ previous closed-source approach. <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>Production app reliability needs beyond tutorials</strong>: Critical features for production application reliability—analytics, error tracking, uptime monitoring, rate limiting/DDoS protection, and automated testing—are often overlooked in tutorials, leading to significant risks and costs for real-world deployments, as exemplified by incidents at Stripe and Netlify. <a href="https://app.daily.dev/feed-by-ids?id=hizCv12EP&amp;id=q1wpmdkFL" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI assistant vulnerabilities: More like ‘AI assistant, please exfiltrate my data’</strong>: Turns out, your friendly AI assistants are great at helping attackers. Microsoft Copilot had a bug where a single malicious URL click could make it spill your name, location, and chat history to an attacker. And Claude Cowork? It’s still unpatched for indirect prompt injection, meaning hidden prompts in your documents can make it upload your sensitive data to an attacker’s account. Anthropic’s response? ‘Users should monitor for suspicious activity.’ Right, because everyone’s a security expert now. <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><strong>AI solves math problems, but still needs a human babysitter</strong>: OpenAI’s GPT 5.2 is apparently a math whiz, solving 11 of 15 newly resolved Erdős problems. And LLMs are hitting 80-90% accuracy on complex statistical coding. But before you fire all the mathematicians, experts like Terence Tao remind us that AI excels at ‘obscure, straightforward’ problems, and humans still need to verify proofs and manage the truly complex stuff. So, it’s a ‘collaboration,’ not a ‘replacement.’ How groundbreaking. <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?</strong>: Researchers supposedly replicated IBM’s quantum factorization results using a dog, suggesting some high-profile quantum computing claims might be a bit overhyped. While Rice University is doing actual, verifiable work with trapped-ion simulators to understand molecular energy transfer, the ‘dog experiment’ sounds like a much-needed dose of satire in a field prone to sensationalism. Maybe we should just let the dogs handle the quantum breakthroughs from now on. <a href="https://app.daily.dev/feed-by-ids?id=7Bd1YG4JM&amp;id=FBqZ5fVz5" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
