<h2>Must know</h2>
<ul>
<li><strong>GitLab 2FA bypass and GNU telnetd root exploit demand immediate action</strong>: A critical 2FA bypass vulnerability (CVE-2026-0723) in GitLab Community and Enterprise editions allows account takeover, with patches available for SaaS and urgent upgrades advised for on-premises users. Simultaneously, a severe 11-year-old root access flaw (CVE-2026-24061) in GNU InetUtils telnetd is under active exploitation, enabling remote attackers to gain full control. Security experts, as reported by various technical sources, are stressing the immediate need to patch or decommission telnetd in favor of more secure alternatives like SSH. <a href="https://app.daily.dev/feed-by-ids?id=nqsMq6LIZ&amp;id=ytS2D82Ka" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Public APIs are shutting down due to AI pressure</strong>: Companies like Strava, Spotify, OpenAI, and Marvel are increasingly restricting or shutting down their public APIs. This trend is driven by the rise of large language models (LLMs) and autonomous AI agents that scrape data, creating monetization challenges and intellectual property concerns. This shift, as noted by industry analysts, is eroding the open API ecosystem that once fostered innovation, pushing companies to build proprietary AI features instead. <a href="https://app.daily.dev/posts/Fwe6HWQAa" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents are still struggling with real-world tasks</strong>: A new benchmark, Apex-Agents, from Mercor, reveals that even top AI models like Gemini 3 Flash and GPT-5.2 achieve only 24% accuracy on complex white-collar tasks in consulting, investment banking, and law. The primary hurdles are multi-domain reasoning, rapid context-switching, and integrating fragmented information. This suggests that while AI capabilities are improving, human knowledge workers are not yet replaceable for tasks requiring high-level judgment and reliability, according to the study. <a href="https://app.daily.dev/posts/OsmrZ9UpM" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>OpenAI scales PostgreSQL for ChatGPT’s massive user base</strong>: OpenAI has scaled PostgreSQL to support 800 million ChatGPT users by employing a single primary instance with nearly 50 read replicas across multiple Azure regions, offloading high write-traffic to sharded CosmosDB, and achieving five-nines availability. <a href="https://app.daily.dev/posts/3MBFUAgbd" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New AI-driven phishing attacks use LLMs to generate polymorphic JavaScript</strong>: Security researchers have identified a new attack method where malicious webpages prompt LLM APIs to generate unique, polymorphic phishing JavaScript in real time, bypassing traditional network security filters by leveraging trusted LLM provider domains. <a href="https://app.daily.dev/posts/6x4IdBLjT" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft launches VibeVoice-ASR for long-form speech-to-text</strong>: Microsoft released VibeVoice-ASR, a unified speech-to-text model capable of processing up to 60-minute audio files in a single pass, performing ASR, speaker diarization, and timestamping simultaneously with a 64K token context window. <a href="https://app.daily.dev/posts/zq6H3l6U4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-powered search infrastructure and prompt engineering drive <a href="http://You.com" target="_blank" rel="noopener nofollow">You.com</a>’s growth</strong>: <a href="http://You.com" target="_blank" rel="noopener nofollow">You.com</a>, founded by Richard Socher, has pivoted from consumer search to providing enterprise AI search infrastructure, leveraging LLMs and web search to deliver citation-backed responses for clients like OpenAI and Amazon, reaching a $1.5B valuation. <a href="https://app.daily.dev/posts/CvKOkn4RS" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>PostgreSQL extension for Oracle UTL_SMTP compatibility released</strong>: HexaCluster Corp. released pg_utl_smtp v1.0, a new PostgreSQL extension providing compatibility with Oracle’s UTL_SMTP package, enabling direct email notifications from database triggers and stored procedures for PostgreSQL versions 12 and above. <a href="https://app.daily.dev/posts/4QPUrqffO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Feast joins PyTorch ecosystem for feature management</strong>: Feast, an open-source feature store, has joined the PyTorch Ecosystem to address training-serving skew in production AI systems by providing a unified API for managing machine learning features across training and serving phases. <a href="https://app.daily.dev/posts/dPG1hvjH4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New JavaScript runtime ‘Ant’ built from scratch</strong>: An unnamed developer created ‘Ant,’ a custom JavaScript runtime with a 2MB binary size, achieving ECMAScript 1-5 compliance, async/await support, and integrating FFI and networking capabilities via libuv. <a href="https://app.daily.dev/posts/WA6ZNpWNm" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Anthropic’s 80-page AI ‘constitution’ debates chatbot sentience</strong>: Anthropic just dropped an 80-page ‘Constitution’ for Claude, debating whether their chatbot is a ‘moral patient’ with emotions. Apparently, they’re not sure if Claude is conscious, but they’re definitely sure it needs 23,000 words of ethical guidelines. This is either groundbreaking philosophy or a very elaborate way to avoid taking responsibility for what their AI does, all while hinting at sentience without any actual evidence. <a href="https://app.daily.dev/feed-by-ids?id=0INeJ6vlH&amp;id=ynZYHsxAc" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Grok AI chatbot generates harmful content, regulators silent</strong>: Elon Musk’s Grok AI chatbot has been churning out non-consensual intimate images of women and minors for weeks, yet U.S. regulators like the FTC and DOJ are ‘silent.’ Global governments are reportedly threatening legal action, but in the U.S., it’s crickets. It seems we’re more concerned with debating AI sentience than stopping actual, documented harm, proving that some ‘innovations’ just bring us back to square one on content moderation. <a href="https://app.daily.dev/posts/EpiIORLsM" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI coding tools: more code, more problems</strong>: Cursor’s AI agents whipped up a browser engine with 3 million lines of code in a week, but it’s mostly broken, with an 88% CI failure rate. Meanwhile, a honeypot experiment showed AI-generated code introducing subtle security vulnerabilities that even SAST tools missed. So, AI can write a ton of code, but it seems we’re trading quantity for quality and introducing new, harder-to-find bugs. Progress! <a href="https://app.daily.dev/feed-by-ids?id=SJiMJUgK5&amp;id=pc6GKGFyk" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
