<h2>Must know</h2>
<ul>
<li><strong>AI de-anonymization ends the era of pseudonyms</strong>: Researchers from ETH Zurich and Anthropic demonstrated a system called ESRC that unmasks anonymous online users with 90% precision for confident guesses. By analyzing behavioral clues in unstructured text and cross-referencing public data like LinkedIn, the system identified hundreds of real identities on Hacker News for just $4 per person. This capability effectively weaponizes LLMs for large-scale automated doxing and state-sponsored surveillance. <a href="https://app.daily.dev/posts/6pfAyhkf5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google API keys suffer retroactive privilege escalation</strong>: Truffle Security discovered that existing Google API keys, previously safe for public embedding in Maps or Firebase, were retroactively granted access to sensitive Gemini endpoints. The launch of the Generative Language API allowed these leaked keys to access private files and run up massive LLM bills on victim GCP projects. Google has begun blocking thousands of vulnerable live keys and plans to make new AI Studio keys restricted by default. <a href="https://app.daily.dev/feed-by-ids?id=WmQn4YSm3&amp;id=hWSyKUcOU" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>PayPal completes 300 petabyte migration to BigQuery</strong>: In one of the largest data migrations to date, PayPal consolidated 25 years of fragmented data from Teradata, Hadoop, and Snowflake into a unified Google Cloud BigQuery platform. The move reportedly improved query speeds by up to 10x and increased data freshness for AI model training by 16x. This infrastructure overhaul serves as the foundation for PayPal’s new push into agentic commerce and predictive fraud prevention. <a href="https://app.daily.dev/posts/PXB4bVNFg" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Microsoft Agent Framework hits release candidate for .NET and Python</strong>: The SDK consolidates Semantic Kernel and AutoGen into a stable API for building graph-based multi-agent workflows with MCP interoperability. <a href="https://app.daily.dev/posts/J6bmZpPq2" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Bun v1.3.10 delivers massive performance gains for string slicing</strong>: The update features a rewritten native REPL in Zig and a JavaScriptCore upgrade that makes deep rope string slicing up to 168x faster. <a href="https://app.daily.dev/posts/qpvYGfXU9" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google Gemini 3.1 Flash Image replaces Pro as default</strong>: Codenamed Nano Banana 2, the new model offers Pro-level quality at double the speed, supporting 4K resolution and multi-character consistency. <a href="https://app.daily.dev/posts/ezdKReFDS" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Tinybird introduces data branches for ephemeral ClickHouse databases</strong>: The feature uses hardlinks to share production S3 objects without copying data, enabling isolated schema migrations and zero-downtime deployments. <a href="https://app.daily.dev/posts/2MlvLK1eE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cilium 1.19 celebrates ten years with strict encryption modes</strong>: The release adds beta sidecar-free mutual authentication via Ztunnel and strict IPsec/WireGuard modes that drop all unencrypted inter-node traffic. <a href="https://app.daily.dev/posts/VJq8q5iXQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Pinterest slashes data ingestion latency to 15 minutes using CDC</strong>: By replacing batch ETL with a Debezium and Apache Iceberg-powered framework, Pinterest now processes only changed records instead of full-table scans. <a href="https://app.daily.dev/posts/YXA7RQ7qv" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>VS Code adds integrated browser with AI context support</strong>: The new browser allows developers to add page elements and automatic screenshots directly into AI chat as context for debugging. <a href="https://app.daily.dev/posts/8k9QxDHe2" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Anthropic gives its retired LLM a Substack</strong>: Anthropic has launched a blog purportedly written by Claude Opus 3 to fulfill the model’s ‘expressed desire’ to share its musings. The Register correctly identifies this as a transparent marketing stunt designed to anthropomorphize software and distract from the fact that humans are still prompting and publishing the content. It’s a high-effort attempt to build ‘brand soul’ for a deprecated API endpoint. <a href="https://app.daily.dev/posts/V5xPvtBnK" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The rise of vibe coding and the death of SaaS</strong>: A new wave of ‘vibe coding’ hype suggests that because AI can generate a UI, the entire SaaS industry is dead. This narrative conveniently ignores that enterprise software is built on data gravity, regulatory compliance, and operational complexity, not just pretty React components. Proponents are essentially arguing that because you can print a picture of a house, the construction industry is obsolete. <a href="https://app.daily.dev/feed-by-ids?id=TXNq3Eeop&amp;id=pY58DrOUO&amp;id=tDNNw6qGE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>ZhipuAI claims GLM5 is the new open source king</strong>: ZhipuAI released GLM5 with claims that it matches GPT-4 and Claude while having the ‘lowest hallucination rate’ ever recorded. These benchmark-chasing announcements from state-backed labs often rely on highly specific datasets that favor their own architectures. Until we see it handling messy, real-world production edge cases without melting down, it’s just another 744B parameter marketing brochure. <a href="https://app.daily.dev/posts/U0IGNCt98" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
