<h2>Must know</h2>
<ul>
<li><strong>AI accelerates exploit and malware development</strong>: LLMs like Opus 4.5 and GPT-5.2 are now generating sophisticated exploits for zero-day vulnerabilities, bypassing modern security mitigations like ASLR and CFI. According to Hacker News, GPT-5.2 solved a complex exploit scenario in just three hours for about $50. Check Point Research also identified VoidLink, the first AI-generated sophisticated malware suite, developed in under a week by a single person using AI agents. This marks a significant shift towards industrialized offensive cybersecurity, where exploit generation is limited by LLM token throughput rather than human expertise. <a href="https://app.daily.dev/feed-by-ids?id=WkT9zuzY1&amp;id=nLTC7GTyu" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kubernetes is now the standard for enterprise AI</strong>: The 2025 CNCF Annual Cloud Native Survey reveals Kubernetes production usage has hit 82%, with 66% of AI adopters using it for inference. This establishes Kubernetes as the go-to infrastructure for enterprise AI workloads. Interestingly, the survey, widely cited by sources, indicates that cultural resistance, not technical complexity, is now the primary barrier to adoption. Mature organizations are leveraging GitOps and platform engineering for scaling AI on Kubernetes. <a href="https://app.daily.dev/posts/Lwsgn3lZ5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic’s Git server vulnerabilities exposed RCE risks</strong>: Anthropic’s Git MCP server had three critical vulnerabilities, including path validation bypass and argument injection, which could lead to remote code execution via prompt injection. These flaws, discovered in June 2025 and patched by December, highlight new attack surfaces in AI-integrated systems. The exploits chained vulnerabilities with the Filesystem MCP server, using Git’s smudge and clean filters to run arbitrary code, demonstrating how AI agent tools can introduce unexpected security risks. <a href="https://app.daily.dev/posts/U6A8HrCeq" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>AI agent architectures are standardizing</strong>: The A2A Java SDK 1.0.0.Alpha1 aligns with the forthcoming Agent2Agent Protocol 1.0, standardizing communication between AI agents, while a secondary article analyzes the division between AI agents and modular Agent Skills to address prompt bloat and scalability issues. <a href="https://app.daily.dev/feed-by-ids?id=LmGT7Q5vx&amp;id=PMa93QJZs&amp;id=ZcWzztsx0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New AI coding model GLM-4.7 Flash outperforms competitors</strong>: GLM-4.7 Flash, a 30B-A3B mixture-of-experts language model, significantly outperforms competitors like Qwen 3-30B-A3B in coding and agentic AI benchmarks, scoring 59.2% on SWBench and 79.5% on Tau Squared Bench, and is available for self-hosting with an MIT license. <a href="https://app.daily.dev/posts/eWdzokgG0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NIST updates AI cybersecurity frameworks</strong>: NIST’s latest AI cybersecurity frameworks (CSF 2.0 and AI Risk Management Framework) emphasize integrating AI security into mainstream cybersecurity operations, focusing on data trust, continuous visibility, and proactive governance to prevent AI from amplifying existing security weaknesses. <a href="https://app.daily.dev/feed-by-ids?id=8uzQJ4GD3&amp;id=QaJ2IQUdX&amp;id=j6bBmLvCW" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google Gemini prompt injection exploit raises enterprise concerns</strong>: Security researchers found a vulnerability in Google Gemini where malicious calendar invitations could exploit the model’s automatic data ingestion, allowing attackers to manipulate its behavior and exfiltrate private meeting data, highlighting the risks of prompt injection in LLM-based systems. <a href="https://app.daily.dev/posts/dHjNBMkFO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft mandates Intune security updates for mobile apps</strong>: Microsoft is enforcing mandatory Intune Mobile Application Management security updates from January 19, 2026, requiring iOS and Android apps to update to the latest Intune SDK versions or face immediate blocking from corporate resources, impacting apps like Outlook and Teams. <a href="https://app.daily.dev/posts/PNEoUht10" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New EU decentralized vulnerability database launched</strong>: The GCVE has launched <a href="http://db.gcve.eu" target="_blank" rel="noopener nofollow">db.gcve.eu</a>, a publicly accessible, decentralized cybersecurity vulnerability database integrating data from over 25 sources, aiming to reduce European reliance on US-based systems like CVE and provide autonomous vulnerability identifier assignment. <a href="https://app.daily.dev/posts/i2uCJ3RGE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Taiwan and US sign AI and semiconductor supply chain pact</strong>: Taiwan and the United States finalized a strategic trade agreement focused on semiconductors and AI supply chains, with Taiwan committing $250 billion in investments and credit guarantees for US production, and Taiwanese chip makers like TSMC receiving lowered tariffs. <a href="https://app.daily.dev/posts/suaVSewS2" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI coding agents: 100x velocity or 100x headache?</strong>: Apparently, AI coding agents like Claude Code can deliver ‘10-100x velocity gains’ for ‘well-defined’ tasks. But then, they ‘perform poorly on ambiguous, high-level product requirements’ and require ‘careful human task decomposition.’ So, they’re great if you already know exactly what you want and how to build it, which kind of defeats the purpose of ‘agentic’ AI, doesn’t it? And let’s not forget the ‘322% more code vulnerabilities’ from tools like GitHub Copilot. Sounds like a fantastic way to accelerate technical debt. <a href="https://app.daily.dev/feed-by-ids?id=EqMZfdLoq&amp;id=GPLZOgOgP&amp;id=J24eplQdr&amp;id=cSbwL1BS5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vibecoding: the new executive superpower?</strong>: Emergent, an Indian AI startup, just bagged $70 million for ‘vibe-coding,’ claiming 5 million users and $50 million ARR. And now, we’re told ‘every exec should be vibecoding’ to build apps without ‘full dependency on engineering staff.’ Because nothing says ‘robust, scalable enterprise solution’ like a CEO ‘vibecoding’ a hackathon platform. I’m sure the engineering teams are thrilled about this new era of ‘individual contributors’ bypassing their expertise. <a href="https://app.daily.dev/feed-by-ids?id=AllrxKqHW&amp;id=McILqauKu" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-driven ERP: 25% productivity gains, 0% CEO confidence</strong>: We’re being sold on ‘AI-driven ERP’ promising 25% productivity gains and 45% processing time reduction. Yet, PwC’s survey of 4,454 CEOs shows 56% report no revenue increase or cost decrease from AI spending, and CEO confidence is at a five-year low. So, the AI is delivering massive benefits, but nobody’s actually seeing them? Sounds like another case of ‘trust us, the numbers are great, just not in your balance sheet.’ <a href="https://app.daily.dev/feed-by-ids?id=Kov24DV15&amp;id=ogb2Wd6xs&amp;id=rEdiBiFbV" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
