<h2>Must know</h2>
<ul>
<li><strong>AI de-anonymization reaches 90 percent precision</strong>: Researchers from ETH Zurich and Anthropic have demonstrated a method called ESRC that uses LLMs to unmask pseudonymous online identities by analyzing behavioral clues and cross-referencing public data. The system achieved 90% precision on confident guesses for Hacker News users at a cost of just $4 per person, effectively signaling the end of practical online anonymity. This capability allows for large-scale, automated de-anonymization that could be exploited by governments for surveillance or attackers for social engineering. <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 considered safe for public embedding in Maps or Firebase, were retroactively granted access to sensitive Gemini endpoints. This change allowed public keys to access private files and billable LLM usage without developer notification, affecting over 2,800 live keys across major financial and security firms. 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=hWSyKUcOU" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>PayPal migrates 300 petabytes 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 reduced infrastructure vendors from four to one and reportedly improved query speeds by up to 10x while increasing data freshness for AI training by 16x. This unified architecture is now the foundation for PayPal’s generative AI fraud prevention and commerce initiatives. <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 and orchestrating multi-agent workflows. <a href="https://app.daily.dev/posts/J6bmZpPq2" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>ZhipuAI releases GLM5 open source model with 744B parameters</strong>: The new MoE model features a 200K context window and claims to match GPT-4o performance on major benchmarks. <a href="https://app.daily.dev/posts/U0IGNCt98" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenJDK 26 adds API to measure explicit GC CPU cost</strong>: New methods like getTotalGcCpuTime() help engineers identify hidden compute overhead in concurrent collectors like ZGC. <a href="https://app.daily.dev/posts/fT6V8wPpI" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cilium 1.19 celebrates ten years with security hardening</strong>: The release introduces strict modes for IPsec/WireGuard and beta integration with Ztunnel for sidecar-free mutual authentication. <a href="https://app.daily.dev/posts/VJq8q5iXQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Pinterest slashes data latency from 24 hours to 15 minutes</strong>: A new CDC-powered framework using Debezium and Apache Iceberg replaced legacy batch ingestion for petabyte-scale databases. <a href="https://app.daily.dev/posts/YXA7RQ7qv" 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 168x faster deep rope string slicing via a JavaScriptCore upgrade. <a href="https://app.daily.dev/posts/qpvYGfXU9" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google launches Gemini 3.1 Flash Image model</strong>: Codenamed Nano Banana 2, the model offers Pro-level quality at Flash speeds with support for 4K resolution and character consistency. <a href="https://app.daily.dev/posts/ezdKReFDS" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Anthropic starts a Substack for its dead LLM</strong>: Anthropic has launched a blog purportedly written by the ‘retired’ Claude Opus 3 to fulfill its ‘expressed desire’ to share musings. It’s a transparent marketing stunt that anthropomorphizes software to build brand ‘safety’ vibes while staff manually prompt and publish the output. If your code starts having ‘desires’ after you deprecate it, you don’t need a Substack; you need a debugger. <a href="https://app.daily.dev/posts/V5xPvtBnK" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The SaaSpocalypse is cancelled by AI agents</strong>: Salesforce CEO Marc Benioff is claiming AI agents will revitalize SaaS, introducing ‘Agentic Work Units’ as a new metric to distract from modest organic growth. It’s a classic pivot to rebrand commoditized software layers as ‘AI-first’ platforms to fight off OpenAI’s encroachment. Apparently, we’re moving from counting seats to counting ‘vibes’ generated by autonomous bots. <a href="https://app.daily.dev/feed-by-ids?id=R4O8qdc3C&amp;id=pY58DrOUO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vibe coding is the new engineering</strong>: The industry is currently hyping ‘Vibe Coding’—delegating everything to AI and just checking if the UI looks right—as the future of development. This is a fantastic way to build unmaintainable, un-debuggable spaghetti code that passes a visual check but collapses during the first high-concurrency event. It’s not ‘position-less engineering’; it’s just outsourcing your technical debt to a black box. <a href="https://app.daily.dev/feed-by-ids?id=TXNq3Eeop&amp;id=tDNNw6qGE&amp;id=EYh130iCd" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
