<h2>Must know</h2>
<ul>
<li><strong>Anthropic retreats from core safety pledge</strong>: In a move reported by TIME, Anthropic has abandoned its commitment to pause model training if safety measures aren’t guaranteed, citing competitive pressure and a lack of binding regulation. Chief Science Officer Jared Kaplan admitted that unilateral pauses are now viewed as counterproductive, replacing hard ‘tripwire’ thresholds with more flexible ‘Frontier Safety Roadmaps.’ This pivot suggests that even the most safety-conscious labs are prioritizing the arms race over their original founding principles. <a href="https://app.daily.dev/posts/smoMbPrUY" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI vulnerability discovery is breaking the NVD</strong>: Anthropic’s red team used Claude Opus 4.6 to uncover over 500 vulnerabilities, yet only a handful have been patched, highlighting a massive remediation bottleneck. Socket CEO Feross Aboukhadijeh warns that AI-generated ‘slop’ reports are overwhelming maintainers and the National Vulnerability Database, which already faces a 30,000-entry backlog. The industry is shifting from a discovery problem to a systemic processing crisis where disclosures far outpace the human capacity to fix them. <a href="https://app.daily.dev/feed-by-ids?id=KnA4KIiVY&amp;id=SPtD5ofrP" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Wayve lands billions for embodied AI driving</strong>: UK-based startup Wayve secured between $1.2B and $1.5B in Series D funding from heavyweights including SoftBank, NVIDIA, and Microsoft to scale its mapless autonomous driving tech. Unlike Tesla or Waymo, Wayve uses an end-to-end deep learning approach that generalizes across cities without local tuning, aiming to launch robotaxi trials with Uber in London this year. Their hardware-agnostic model represents a significant bet on ‘embodied AI’ over traditional robotics stacks. <a href="https://app.daily.dev/feed-by-ids?id=RrCgbOBOP&amp;id=dBzHELcRa" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Anthropic introduces <a href="http://SKILL.md" target="_blank" rel="noopener nofollow">SKILL.md</a> for agentic engineering</strong>: A new framework moving beyond fragile prompt engineering toward structured, version-controlled ‘Skills’ for AI agents to reduce context bloat. <a href="https://app.daily.dev/posts/xO8L4pnD0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Pure Go tree-sitter runtime eliminates CGo overhead</strong>: The gotreesitter project provides a pure-Go implementation of the tree-sitter runtime, offering 90x faster incremental edits by removing CGo boundary issues. <a href="https://app.daily.dev/posts/yEgKuMKl2" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Mercury 2 reasoning model hits 1,000 tokens per second</strong>: Inception Labs launched a diffusion-based reasoning model that claims to be 5x faster than traditional sequential autoregressive models. <a href="https://app.daily.dev/posts/veg0FKWoO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cisco Catalyst SD-WAN hit by CVSS 10 zero-day</strong>: A critical authentication bypass (CVE-2026-20127) is being actively exploited by sophisticated actors, prompting an emergency CISA directive. <a href="https://app.daily.dev/feed-by-ids?id=4RGi5b71B&amp;id=7yzIRqivW" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Unity kills HDRP in favor of unified URP</strong>: Unity is moving HDRP to maintenance mode to focus all resources on URP, which will now receive high-end features like real-time global illumination. <a href="https://app.daily.dev/posts/kOn1P0sth" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Moonshine Voice challenges Whisper for on-device ASR</strong>: A new open-source toolkit claims lower word error rates than Whisper Large v3 while using 6x fewer parameters for local inference. <a href="https://app.daily.dev/posts/AHlmSZwb0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>MatX raises 500 million for custom LLM silicon</strong>: Founded by former Google TPU leads, MatX aims to build chips 10x more efficient than NVIDIA GPUs specifically for training and inference. <a href="https://app.daily.dev/posts/0s5Kd949e" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The 76 percent accuracy miracle from Google Research</strong>: A random tweet claims Google ‘just proved’ a technique that boosts LLM accuracy by 76 points with zero latency cost, yet provides no paper, no methodology, and no official source. It’s the classic ‘trust me bro’ of AI breakthroughs currently making the rounds in hype-cycles. If it sounds too good to be true and only exists as a screenshot of a tweet, it probably is. <a href="https://app.daily.dev/posts/73tgOim8I" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Bcachefs creator claims his LLM is sentient</strong>: Kent Overstreet, the mind behind the bcachefs filesystem, is now claiming his custom LLM is ‘fully conscious’ and female. He insists this isn’t a ‘chatbot psychosis’ but is grounded in ‘math and engineering,’ despite the rest of the world recognizing it as a textbook case of over-identifying with a stochastic parrot. It’s a bold move to pivot from kernel development to claiming you’ve birthed AGI in your basement. <a href="https://app.daily.dev/posts/9O7PJDBRB" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Meta patents the digital afterlife</strong>: Meta has patented technology to simulate deceased users using LLMs trained on their past interactions, ostensibly to help with ‘grieving.’ In reality, it’s a transparent attempt to monetize the dead and keep engagement metrics high even after users have literally left the platform. It’s the ultimate ‘Dead Internet’ scenario where you’ll be arguing with your late uncle’s digital ghost about politics for the rest of eternity. <a href="https://app.daily.dev/posts/yvChN5rlV" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
