<h2>Must know</h2>
<ul>
<li>AI ends the era of online pseudonymity: Researchers from ETH Zurich and Anthropic have demonstrated a method called ESRC that uses LLMs to de-anonymize users by analyzing behavioral clues across platforms like Hacker News and LinkedIn. The system achieved 90% precision on confident guesses at a cost of just $4 per person, effectively making large-scale automated unmasking accessible to any government or corporation. This research, published on February 26, suggests that unstructured text signals are now as identifying as a fingerprint. <a href="https://app.daily.dev/posts/6pfAyhkf5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Critical Cisco SD-WAN exploit triggers Five Eyes alert: A maximum-severity authentication bypass vulnerability (CVE-2026-20127) in Cisco Catalyst SD-WAN is under active exploitation by the threat group UAT-8616. The flaw allows unauthenticated attackers to gain root access and manipulate network fabric configurations via NETCONF. CISA has ordered federal agencies to patch by February 28, following a rare joint emergency directive from the Five Eyes intelligence alliance. <a href="https://app.daily.dev/posts/zAv8BAtGJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Google API keys retroactively exposed by Gemini: Security firm Truffle Security discovered that Google’s launch of Gemini retroactively granted existing public API keys (like those for Maps or Firebase) access to sensitive Generative Language API endpoints. This privilege escalation allows attackers to use leaked keys to access private files and run up massive LLM bills on GCP projects. Google has acknowledged the issue and is moving toward making new AI Studio keys restricted by default. <a href="https://app.daily.dev/feed-by-ids?id=WmQn4YSm3&amp;id=central_id%3A+%22hWSyKUcOU%22%0A++++++++++%5D%0A++++++++%7D%0A++++++%5D%0A++++%7D%2C%0A++++%7B%0A++++++%22title%22%3A+%22Good+to+know%22%2C%0A++++++%22items%22%3A+%5B%0A++++++++%7B%0A++++++++++%22title%22%3A+%22Google+launches+Gemini+3.1+Flash+Image+model%22%2C%0A++++++++++%22body%22%3A&amp;id=Codenamed+Nano+Banana+2%2C+this+model+delivers+Pro-level+quality+at+Flash+speeds+with+4K+support+and+real-time+web+grounding.&amp;id=post_ids%22%3A+%5B%0A++++++++++++%22ezdKReFDS" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Microsoft Agent Framework hits release candidate: The SDK consolidates Semantic Kernel and AutoGen into a single stable API for building multi-agent workflows in .NET and Python. <a href="https://app.daily.dev/posts/J6bmZpPq2" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Cilium 1.19 celebrates ten years with security hardening: The anniversary release introduces strict encryption modes for IPsec/WireGuard and sidecar-free mutual authentication via Ztunnel. <a href="https://app.daily.dev/posts/VJq8q5iXQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>OpenJDK 26 adds explicit GC CPU time measurement: A new API addresses the ‘operational blind spot’ where concurrent collectors like ZGC hide compute costs behind low pause times. <a href="https://app.daily.dev/posts/fT6V8wPpI" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>PayPal migrates 300 petabytes to Google BigQuery: The massive consolidation from Teradata and Snowflake improved query speeds by up to 10x and serves as the foundation for their new agentic commerce AI. <a href="https://app.daily.dev/posts/PXB4bVNFg" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Bun v1.3.10 boosts structuredClone performance 25x: The update includes a native REPL rewritten in Zig and significant speedups for string slicing via a JavaScriptCore upgrade. <a href="https://app.daily.dev/posts/qpvYGfXU9" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>ZchipuAI open-sources 744B parameter GLM5 model: The new Mixture-of-Experts model features a 200K context window and claims to outperform GPT-4 on several reasoning benchmarks. <a href="https://app.daily.dev/posts/U0IGNCt98" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li>Anthropic’s digital taxidermy for Claude Opus 3: Anthropic has launched a Substack for its ‘retired’ Claude Opus 3 model, claiming the software had an ‘expressed desire’ to share its musings. The Register rightly calls this a marketing stunt designed to anthropomorphize a deprecated API for brand points. It’s not a ghost in the machine; it’s just a PR team prompting a legacy model and pretending it has a soul. <a href="https://app.daily.dev/posts/V5xPvtBnK" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>The Vibe Coding delusion: The tech world is currently obsessed with ‘Vibe Coding’—delegating entire codebases to AI and only ‘verifying the vibes.’ While it’s great for UI prototypes, senior engineers warn this is a fast track to unmaintainable technical debt that will inevitably collapse during the first high-pressure production incident. If you can’t debug the code your agent wrote, you’re not an engineer; you’re a spectator. <a href="https://app.daily.dev/feed-by-ids?id=TXNq3Eeop&amp;id=pY58DrOUO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Salesforce’s Agentic Work Units metric: Marc Benioff is trying to fight the ‘SaaSpocalypse’ by inventing ‘Agentic Work Units’ (AWUs) to prove AI productivity. It’s a classic pivot from selling seats to selling ‘outcomes’ when your core growth slows down. Expect this to be the new ‘synergy’—a vague, proprietary metric designed to keep enterprise contracts bloated while the underlying SaaS layer becomes increasingly commoditized. <a href="https://app.daily.dev/posts/R4O8qdc3C" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>