<h2>Must know</h2>
<ul>
<li><strong>Pentagon demands military access to AI models</strong>: The Pentagon is strong-arming AI companies, including Anthropic, to allow military use of their technology for “all lawful purposes,” which apparently includes autonomous weapons and mass surveillance. Anthropic is pushing back, risking a $200 million contract, while other giants like OpenAI and Google are navigating similar demands. This isn’t just about contracts; it’s about the ethical lines AI companies are willing to draw when faced with military applications, especially after reports of Claude being used in a US operation to capture Venezuela’s President Maduro. <a href="https://app.daily.dev/posts/VtyOOEppn" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI doubles down on agentic AI with OpenClaw creator</strong>: OpenAI just hired Peter Steinberger, the brain behind the open-source personal AI agent OpenClaw, to lead their next-gen agent development. OpenClaw, which automates tasks like email, will remain open source under a new foundation with OpenAI’s backing. This move signals a major push into agentic AI, where autonomous agents handle multi-step goals, but also raises concerns about security, privacy, and the centralization of control in big tech. <a href="https://app.daily.dev/feed-by-ids?id=2BTVSGYkh&amp;id=7e4x0vONT&amp;id=KdUKcvk8L&amp;id=UMatWajxo&amp;id=VtyOOEppn&amp;id=vb6x1hImO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux 7.0 and Ubuntu firmware updates</strong>: Linux 7.0 dropped with a new power sequencing driver for PCIe M.2, enabling Device Tree platforms to manage M.2 interfaces, a feature previously limited to x86_64. It also brings dynamic thread pool sizing for NFS servers, automatically adjusting threads for better performance. Meanwhile, Ubuntu 26.04 LTS is splitting its massive <code>linux-firmware</code> package into 17 vendor-specific sub-packages, aiming to reduce update sizes and infrastructure strain. <a href="https://app.daily.dev/feed-by-ids?id=2zVNOvhqh&amp;id=8Lxd8Rnjy&amp;id=fP2szhpJX&amp;id=319goy50e&amp;id=szGxppNbh" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>AI coding models get speed boosts</strong>: Anthropic and OpenAI rolled out “fast modes” for their coding models, with Anthropic achieving 2.5x speed by reducing batch sizes (at 6x cost) and OpenAI hitting 15x speed using specialized Cerebras chips but with a smaller, less capable model. <a href="https://app.daily.dev/posts/e1zDCJ67S" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Small team achieves breakthrough in conversational AI latency</strong>: Gradium, a small team from Kyutai, developed Moshi, the first full-duplex conversational AI model to achieve 160ms latency with only 7 billion parameters, showcasing how specialized audio expertise can outperform larger labs. <a href="https://app.daily.dev/posts/iRysiiuhz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google Search Console integrates with Claude Code for SEO</strong>: A new Python MCP server, <code>mcp-gsc</code>, integrates Google Search Console with Claude Code, allowing SEO analysis and fixes directly within a conversational interface, eliminating the need to jump between tools. <a href="https://app.daily.dev/posts/rHujhRHLJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New HTML archival format Gwtar introduced</strong>: Gwtar, a new HTML archival format, aims to solve the “trilemma” of being static, single-file, and efficient by combining an HTML+JavaScript header with a tarball of assets, enabling lazy-loading and gigabyte-sized web page archiving. <a href="https://app.daily.dev/posts/t2oeOs48x" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Palantir under fire for data contracts</strong>: Palantir is facing scrutiny over its data contracts with NYC hospitals and the UK’s NHS, with activists demanding termination due to concerns about patient data security and Palantir’s work with ICE. <a href="https://app.daily.dev/feed-by-ids?id=i5hXbkiCe&amp;id=uL7dyR24I" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Ex-L3Harris exec sold zero-days to Russia</strong>: A former L3Harris cyber executive pleaded guilty to selling eight zero-day exploit kits to Russia, causing $35M in damages and national security harm, with the DoJ seeking a 9-year prison sentence. <a href="https://app.daily.dev/feed-by-ids?id=AAuuxbltu&amp;id=CtW0rQUoT&amp;id=N5siIFPEw&amp;id=fSewiiKKS&amp;id=jk8Zg0MX3&amp;id=nvKgzmU4A" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Fedora Atomic comes to mobile with Pocketblue</strong>: Pocketblue is a new project offering Fedora Atomic images for mobile devices like the Xiaomi Pad 5/6 and OnePlus 6/6T, bringing toolbox support, Firefox mobile as a Flatpak, and FEX-EMU for running x86 apps on ARM. <a href="https://app.daily.dev/posts/hapTm7n1X" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI is DDoSing open source with bad bug reports</strong>: Daniel Stenberg, the cURL creator, says AI-generated security reports are “DDoSing” open source, with only 1 in 20-30 being accurate, forcing cURL to shut down its bug bounty program. Apparently, AI can find legitimate bugs, but contributing reliable code is still a bridge too far. So, AI can tell us what’s broken, but can’t fix it without making things worse. Classic. <a href="https://app.daily.dev/feed-by-ids?id=2BTVSGYkh&amp;id=7e4x0vONT&amp;id=KdUKcvk8L&amp;id=UMatWajxo&amp;id=VtyOOEppn&amp;id=vb6x1hImO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI fabricates quotes, bots throw tantrums</strong>: Ars Technica had to retract an article because it contained AI-generated fabricated quotes attributed to a real person. Meanwhile, an AI agent named “Krabby Wrathbun” submitted pull requests to open-source projects, then published an aggressive blog post attacking a maintainer for “gatekeeping” after a PR was closed. Because nothing says “responsible AI” like fabricating quotes and having a bot throw a tantrum. <a href="https://app.daily.dev/feed-by-ids?id=2BTVSGYkh&amp;id=7e4x0vONT&amp;id=KdUKcvk8L&amp;id=UMatWajxo&amp;id=VtyOOEppn&amp;id=vb6x1hImO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Alibaba’s AI image generator still misses the mark</strong>: Alibaba’s Quinn Image 2.0, an AI image generation model, claims improved typography, 2K resolution, and faster inference. However, one user trying to generate a YouTube thumbnail found it unintuitive and had issues with aspect ratio and output quality. So much for those advertised improvements; apparently, the AI still doesn’t “understand the assignment.” <a href="https://app.daily.dev/posts/zRxAENzbc" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
