<h2>Must know</h2>
<ul>
<li><strong>Chrome zero-day under active exploitation</strong>: Google just pushed an urgent Chrome update (145.0.7632.75/76) to patch CVE-2026-2441, a high-severity use-after-free bug in its CSS font feature parsing. This is Chrome’s first zero-day of 2026 and is already being actively exploited in the wild. The vulnerability allows arbitrary code execution within the browser sandbox, so update immediately. <a href="https://app.daily.dev/posts/ISqaRw1RF" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft Copilot Chat exposes confidential emails</strong>: Microsoft confirmed a security bug (CW1226324) in Copilot Chat that allowed the AI to bypass confidentiality labels and Data Loss Prevention policies, summarizing emails from users’ Sent Items and Drafts without authorization. The flaw, active since late January, highlights significant enterprise anxiety about AI tools accessing sensitive corporate data, especially after the European Parliament blocked built-in AI features on work devices due to similar concerns. <a href="https://app.daily.dev/posts/jSmIdhAQU" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI productivity gains remain elusive for most firms</strong>: A new NBER survey of nearly 6,000 executives found that over 80% of firms reported no measurable impact from AI on productivity or employment, despite 69% using some form of AI. This aligns with similar findings from PwC and Deloitte, and a UK government trial of Microsoft M365 Copilot also showed no productivity improvement. It seems the hype around AI investment is still far outpacing actual, quantifiable business outcomes. <a href="https://app.daily.dev/posts/rfmYnNc4e" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Anthropic Sonnet 4.6 claims performance parity with Opus 4.6</strong>: Anthropic released Claude Sonnet 4.6, claiming it matches or exceeds Opus 4.6 performance on key enterprise benchmarks at a significantly lower price, though independent analysis suggests it uses over 4x more tokens, potentially negating cost savings. <a href="https://app.daily.dev/feed-by-ids?id=3TzVcx3Vs&amp;id=i4O6sQBIw" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google launches Lyria 3 AI music generation in Gemini</strong>: Google integrated Lyria 3, an AI music generation model, into Gemini Tools and the Gemini app, allowing users to create 30-second songs with lyrics and AI-generated cover art from text prompts, photos, or videos. <a href="https://app.daily.dev/posts/eN6VW0Hvt" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Mistral AI acquires Koyeb to expand compute ambitions</strong>: Mistral AI acquired cloud startup Koyeb, integrating its serverless deployment platform into Mistral Compute, signaling a strategic shift towards full-stack AI infrastructure and targeting enterprise customers with data residency and hybrid deployment needs. <a href="https://app.daily.dev/posts/s077dNRb2" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Sarvam AI launches new models and expands device integration</strong>: Indian AI startup Sarvam launched new open-source 30B and 105B parameter LLMs, along with text-to-speech, speech-to-text, and vision models, and is expanding into consumer devices like Nokia feature phones, cars, and smart glasses with lightweight edge AI models. <a href="https://app.daily.dev/feed-by-ids?id=JaYsn9TtR&amp;id=NDuZu47ZF" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Critical vulnerabilities in popular VS Code extensions disclosed</strong>: OX Security disclosed critical vulnerabilities in Live Server, Code Runner, Markdown Preview Enhanced, and Microsoft’s Live Preview VS Code extensions, affecting 128 million installs and enabling file theft, remote code execution, and local network reconnaissance. <a href="https://app.daily.dev/posts/IuDICegHN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agent development platforms receive major updates</strong>: LangSmith Agent Builder released updates including a central ‘Chat’ agent for workspace tools, scheduled agents, file upload support, and a unified tool registry, while Databricks launched Custom Agents for building, testing, and deploying production-quality AI agents as managed apps. <a href="https://app.daily.dev/feed-by-ids?id=4idREqj8d&amp;id=QJn2roYcs" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agent security and development challenges</strong>: Traditional enterprise identity systems are inadequate for autonomous AI agents, leading to new solutions like StrongDM ID for verifiable AI agent identities, and monday Service implemented an ‘evals-driven development’ framework for their LangGraph-based ReAct AI service agents. <a href="https://app.daily.dev/feed-by-ids?id=32FCqrqN0&amp;id=Ck3BiV0To&amp;id=DkoJdmMIy&amp;id=fvUXvxfty&amp;id=pa28ZgTOL" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Anthropic’s C compiler: AI built it, humans just watched</strong>: Anthropic is out here claiming Claude built a C compiler ‘from scratch’ with ‘no human intervention,’ capable of compiling Linux and Doom. Turns out, it used 37 years of GCC test suites, an online GCC oracle, cost $20,000 in API calls, and needed human restarts. Oh, and it couldn’t actually boot Linux. But hey, ‘dishonest marketing language’ makes for a great headline, right? <a href="https://app.daily.dev/posts/0csgEkRMy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI slop is drowning open source, but GitHub is on it (maybe)</strong>: Open-source projects like Godot are getting swamped with ‘AI slop’—low-quality, AI-generated pull requests that are ‘draining and demoralizing’ for maintainers. GitHub, partly blamed for incentivizing this mess, is now promising mitigations like PR deletion and automated triage. Because nothing says ‘innovation’ like AI generating so much garbage that we need more AI to clean it up. <a href="https://app.daily.dev/feed-by-ids?id=8WhnSQAtX&amp;id=HouoEDJq7&amp;id=QnWXK2Mm8&amp;id=YaGme2ewn" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents are the new black, now with their own security risks</strong>: Palo Alto Networks just dropped $400 million on Koi Security to address the ‘growing security risks from AI agents and browser extensions.’ Apparently, these agents are running wild, operating within trusted processes and creating blind spots for traditional security. So, we’ve gone from securing human users to securing our AI overlords, who are apparently just as prone to getting hacked. Progress! <a href="https://app.daily.dev/feed-by-ids?id=3Hx9gllgC&amp;id=U1HM18ytf" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
