<h2>Must know</h2>
<ul>
<li><strong>Google Chrome zero-day under active exploitation</strong>: Google has released an urgent update for Chrome (version 145.0.7632.75/76) to patch CVE-2026-2441, a high-severity use-after-free vulnerability in its CSS font feature parsing. This is Chrome’s first zero-day of 2026 and is actively being exploited in the wild. The flaw allows arbitrary code execution within the browser sandbox, making immediate updates critical for all users. <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>: A security bug (CW1226324) in Microsoft Copilot Chat allowed the AI to bypass confidentiality labels and Data Loss Prevention (DLP) policies, summarizing emails from users’ Sent Items and Drafts folders without authorization. The flaw, active since late January, highlights enterprise anxiety about AI tools accessing sensitive corporate data, with Microsoft rolling out a fix but not disclosing the number of affected customers. <a href="https://app.daily.dev/posts/jSmIdhAQU" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Chinese cyberespionage group exploits Dell zero-day for 18 months</strong>: A Chinese cyberespionage group, UNC6201, exploited a critical zero-day vulnerability (CVE-2026-22769) in Dell RecoverPoint for Virtual Machines for approximately 18 months. The flaw, stemming from hardcoded admin credentials in Apache Tomcat Manager, allowed unauthenticated root-level command execution. Mandiant researchers discovered the vulnerability, which has now been patched by Dell. <a href="https://app.daily.dev/posts/TLmbAU6Uq" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>AI productivity gains remain elusive for most businesses</strong>: A recent NBER survey of nearly 6,000 executives found over 80% reported no measurable impact from AI on productivity or employment, despite 69% using some form of AI. This aligns with other surveys and a UK government trial of Microsoft M365 Copilot, suggesting a widening gap between AI investment hype and actual business outcomes. <a href="https://app.daily.dev/posts/rfmYnNc4e" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Mistral AI acquires Koyeb to expand compute ambitions</strong>: Mistral AI has acquired cloud startup Koyeb, integrating its serverless deployment platform into Mistral Compute. This first acquisition signals Mistral AI’s strategic shift towards full-stack AI infrastructure, aiming to target enterprise customers with data residency and hybrid deployment needs, positioning itself as a potential sovereign European alternative to US hyperscalers. <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 has launched new open-source 30B and 105B parameter LLMs, along with text-to-speech, speech-to-text, and vision models. The company is expanding its reach beyond enterprise, deploying lightweight edge AI models on consumer devices like Nokia feature phones, cars, and smart glasses, aiming for offline execution in regions with limited connectivity. <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 found in popular VS Code extensions</strong>: OX Security disclosed critical vulnerabilities in four popular VS Code extensions—Live Server, Code Runner, Markdown Preview Enhanced, and Microsoft’s Live Preview—collectively accounting for 128 million installs. These flaws enable file theft, remote code execution, and local network reconnaissance, with some maintainers failing to respond to disclosure notices. <a href="https://app.daily.dev/posts/IuDICegHN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Palo Alto Networks acquires Koi Security for AI agent governance</strong>: Palo Alto Networks acquired Koi Security for approximately $400 million, aiming to integrate Koi’s platform for monitoring and governing software components, including AI models and browser extensions, into its Prisma AIRS and Cortex XDR platforms. This move addresses growing security risks from AI agents and non-human identities in corporate environments. <a href="https://app.daily.dev/feed-by-ids?id=3Hx9gllgC&amp;id=U1HM18ytf&amp;id=70bwD2IsO&amp;id=Hx9SCREZp&amp;id=IOwwjUreT&amp;id=L7iJSIIDS&amp;id=VukD3HNtV" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agent ecosystem navigates policy shifts and data challenges</strong>: Anthropic clarified its Claude Code documentation regarding OAuth token usage in third-party tools after community backlash, while OpenAI partnered with Indian fintech Pine Labs to embed AI APIs into its infrastructure. Amazon shared a framework for evaluating agentic AI systems, and Dropbox built a scalable context engine for Dash, highlighting the rapid evolution and challenges in the AI agent ecosystem. <a href="https://app.daily.dev/feed-by-ids?id=1UIQguzNO&amp;id=3lsZ5Vula&amp;id=D45i7bkfM&amp;id=EZOdc00Xc&amp;id=FQxtQgg9R&amp;id=FTWE8YzDG&amp;id=JlixRkBbI&amp;id=JqU9WLooM&amp;id=LKtTxw05m&amp;id=R2tIcO414" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI partners with Tata Group for India expansion</strong>: OpenAI partnered with India’s Tata Group to secure 100MW of AI-ready data center capacity through TCS’s HyperVault business, with plans to scale to 1GW. This strategic move, part of ‘OpenAI for India,’ aims to reduce latency, address data residency, and deploy ChatGPT Enterprise for TCS employees, reflecting India’s growing importance in the AI landscape. <a href="https://app.daily.dev/posts/6fxZc9j6p" 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 marketed it badly</strong>: Anthropic released a marketing video claiming Claude built a C compiler ‘from scratch’ with ‘no human intervention.’ Turns out, it used 37 years of GCC test suites, an online GCC oracle, cost $20,000 in API calls, and needed human restarts. The actual achievement of 16 AI agents autonomously running for two weeks to produce a 100,000-line compiler is impressive, but the ‘dishonest marketing language’ (0csgEkRMy) is just classic AI hype. <a href="https://app.daily.dev/posts/0csgEkRMy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI will make you a better writer, just not with your own thoughts</strong>: Some ‘experts’ are now claiming AI is commoditizing ‘lower-tier writing’ like tutorials and news, while ‘authentic’ writing (you know, the stuff with actual human thought) remains valuable. Apparently, outsourcing your thinking to AI leads to ‘cognitive debt’ and makes you lose touch with your own brain. So, AI is great for writing, just not for writing <em>your</em> writing. Got it. <a href="https://app.daily.dev/feed-by-ids?id=4HNvjLVYH&amp;id=rnFNxSucm" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents are the new black, but still need babysitting</strong>: Everyone’s jumping on the AI agent bandwagon, with new platforms like LangSmith Agent Builder and Databricks Custom Agents promising to make AI agents ‘production-ready.’ Yet, demos often fail in production due to ‘unbounded telemetry data’ and ‘insufficient vector memory.’ So, while the marketing says ‘autonomous,’ the reality is still ‘requires constant human supervision and a lot of duct tape.’ <a href="https://app.daily.dev/feed-by-ids?id=4idREqj8d&amp;id=QJn2roYcs&amp;id=l5kVESwg0&amp;id=qBKhvArRq&amp;id=wLx806hAH" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
