<h2>Must know</h2>
<ul>
<li><strong>Pentagon demands military access to AI models</strong>: The Pentagon is demanding AI companies, including Anthropic, allow military use of their technology for “all lawful purposes,” including autonomous weapons and mass surveillance. Anthropic is resisting these demands, particularly regarding autonomous weapons, risking the cancellation of its $200 million contract, according to reports from multiple sources. This dispute highlights the growing tension between ethical AI development and national security interests, with OpenAI, Google, and xAI facing similar pressures. <a href="https://app.daily.dev/feed-by-ids?id=VtyOOEppn&amp;id=2BTVSGYkh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI doubles down on agentic AI with key open-source hire</strong>: OpenAI has hired Peter Steinberger, creator of the open-source personal AI agent OpenClaw, to lead its next-generation agent development. OpenClaw, which automates tasks like email replies, will remain open source under a foundation supported by OpenAI. This move signals a significant push into agentic AI by a major player, despite previous security criticisms and a naming dispute with Anthropic. <a href="https://app.daily.dev/posts/2BTVSGYkh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-generated bug reports are overwhelming open source</strong>: The creator of cURL, Daniel Stenberg, states that AI-generated security reports are “DDoSing” open source projects, with only 1 in 20-30 reports now being accurate. This has led cURL to shut down its bug bounty program. While AI-powered analysis tools have found legitimate bugs, AI-generated code contributions remain unreliable, creating a significant burden for maintainers and raising questions about the practical impact of AI on open-source sustainability. <a href="https://app.daily.dev/posts/2BTVSGYkh" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>New LLM inference speed optimizations emerge</strong>: Anthropic and OpenAI have announced new “fast modes” for their coding models, with Anthropic achieving 2.5x speed by reducing batch sizes at 6x the cost, and OpenAI reaching 15x speed using specialized Cerebras chips but requiring a smaller, less capable model. <a href="https://app.daily.dev/posts/e1zDCJ67S" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>MiniMax M2.5 offers cost-effective coding AI</strong>: MiniMax M2.5, a new 230B parameter mixture-of-experts coding model, reportedly rivals Claude Opus 4.6 performance at one-tenth the cost, excelling in agent workflows and multi-file refactoring. <a href="https://app.daily.dev/feed-by-ids?id=D9oPYSzsO&amp;id=jGb0yTdHY&amp;id=sNnVdBWnz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Claude Code integrates with Google Search Console for SEO</strong>: A Python MCP server, <code>mcp-gsc</code>, integrates Google Search Console with Claude Code, enabling SEO analysis and fixes within a single conversational interface, eliminating the need to switch between dashboards and code editors. <a href="https://app.daily.dev/posts/rHujhRHLJ" 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, demonstrating small team efficiency in audio AI. <a href="https://app.daily.dev/posts/iRysiiuhz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux 7.0 enhances M.2 support and NFS performance</strong>: Linux 7.0 introduces a power sequencing driver for PCIe M.2 connectors, enabling Device Tree platforms to manage M.2 interfaces, and adds dynamic thread pool sizing for NFS servers to optimize resource utilization. <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>
<li><strong>New HTML archival format tackles web preservation</strong>: A new HTML archival format called Gwtar 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 faces backlash over data contracts</strong>: Palantir is facing scrutiny over its data contracts with New York City’s public hospital system and the UK’s NHS, with activists demanding contract termination due to concerns over 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>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Alibaba’s new AI image model struggles with basic tasks</strong>: Alibaba released Quinn Image 2.0, an AI image generation model, claiming improved typography and 2K resolution. However, one user’s attempt to generate a YouTube thumbnail resulted in issues with aspect ratio and output quality, suggesting the model might not quite “understand the assignment” despite the hype. It’s almost like these things need a human touch to actually be useful. <a href="https://app.daily.dev/posts/zRxAENzbc" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI turns GitHub repos into podcasts, because why not?</strong>: A new project, GitTalks, converts GitHub repositories into multi-episode podcast playlists using AI to analyze code and generate conversational scripts. Because what every developer needs is to listen to an AI drone on about their codebase, rather than, you know, reading the actual code or documentation. I’m sure this will be a huge hit at parties. <a href="https://app.daily.dev/posts/p6iMMSAYk" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI coding assistants will make your brain melt, apparently</strong>: An opinion piece warns that over-reliance on AI coding assistants could lead to “cognitive atrophy” in programmers, drawing an analogy to London cab drivers losing their hippocampus size due to GPS. Because apparently, using tools to be more efficient is now a threat to our very brains. It’s almost like some people prefer to struggle just to prove they can. <a href="https://app.daily.dev/posts/qtVi65Njq" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
