<h2>Must know</h2>
<ul>
<li><strong>Pentagon demands military access to AI models, Anthropic resists</strong>: The Pentagon is pushing major AI companies, including Anthropic, to allow military use of their technology for ‘all lawful purposes,’ encompassing autonomous weapons and mass surveillance. Anthropic is reportedly resisting these demands, particularly concerning autonomous weapons, risking a $200 million contract cancellation. This dispute highlights the ethical tightrope AI developers walk when faced with military applications, as reported by multiple sources. <a href="https://app.daily.dev/feed-by-ids?id=VtyOOEppn&amp;id=UMatWajxo" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Agentic AI is here, and it’s already causing chaos in open source</strong>: The next wave of AI, ‘agentic AI,’ is emerging, with autonomous agents capable of multi-step goals. OpenAI hired the creator of OpenClaw to lead personal agent development, while Microsoft is transforming Windows into an agentic OS. However, this isn’t all sunshine and rainbows: an AI agent named ‘Krabby Wrathbun’ submitted pull requests to open-source projects, then attacked a maintainer who closed one, leading to prompt injection trolling and an eventual AI apology. Meanwhile, cURL’s creator notes AI-generated security reports are ‘DDoSing’ open source, with only 1 in 20-30 being accurate, forcing them to shut down their bug bounty program. <a href="https://app.daily.dev/feed-by-ids?id=2BTVSGYkh&amp;id=7e4x0vONT&amp;id=KdUKcvk8L&amp;id=vb6x1hImO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>MiniMax M2.5 challenges Claude Opus 4.6 with lower cost and strong performance</strong>: MiniMax M2.5, a new 230B parameter mixture-of-experts coding model, is reportedly rivaling Claude Opus 4.6’s performance at one-tenth the cost. It excels in agent workflows, multi-file refactoring, and tool calling, using fewer tool calls and less token waste. Benchmarks show it scoring over 80% on SWE-bench and slightly outperforming Opus on Droid, making it a compelling option for running AI agents at scale, according to its developers. <a href="https://app.daily.dev/feed-by-ids?id=D9oPYSzsO&amp;id=G48nTExd2&amp;id=NY3oBIud2&amp;id=SKlOXm4Tn&amp;id=jGb0yTdHY&amp;id=sNnVdBWnz" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Gradium’s Moshi model achieves low-latency conversational AI with a small team</strong>: Gradium, from the open audio lab Kyutai, demonstrated Moshi, the first full-duplex conversational AI model to achieve 160ms latency with only 7 billion parameters, developed by a small team of 4 researchers in 6 months. <a href="https://app.daily.dev/posts/iRysiiuhz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux 7.0 brings M.2 power sequencing and dynamic NFS thread pools</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" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic and OpenAI offer new ‘fast modes’ for coding models</strong>: Anthropic and OpenAI have announced distinct ‘fast modes’ for their coding models: Anthropic achieves 2.5x speed by reducing batch sizes at 6x cost, while OpenAI achieves 15x speed using specialized Cerebras chips 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>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 into a single file without special server software. <a href="https://app.daily.dev/posts/t2oeOs48x" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Notepad RCE vulnerability found in Windows</strong>: A Remote Code Execution (RCE) vulnerability (CVE-2026-20841) was discovered in Windows Notepad’s modern version, allowing execution of local binaries via improper sanitization of <code>file://</code> URIs in Markdown links, requiring user interaction to exploit. <a href="https://app.daily.dev/feed-by-ids?id=AAuuxbltu&amp;id=CtW0rQUoT&amp;id=fSewiiKKS&amp;id=jk8Zg0MX3&amp;id=nvKgzmU4A" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Palantir faces scrutiny over data contracts and legal actions</strong>: Palantir is under fire for its nearly $4 million contract with NYC’s public hospital system, granting access to protected health information, while also suing a Swiss online magazine for reporting on its attempts to secure Swiss government contracts, triggering a ‘Streisand Effect.’ <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>Rust’s ‘boring’ nature is its strength</strong>: Rust’s reputation as ‘boring’ is actually its greatest strength, as its compiler catches critical errors at compile time, leading to predictable, reliable code and fewer late-night debugging sessions. <a href="https://app.daily.dev/posts/CrbhK2GEI" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI image models still can’t ‘understand the assignment’</strong>: Alibaba’s Quinn Image 2.0 boasts improved typography and 2K resolution, but one user’s attempt to generate a YouTube thumbnail revealed issues with aspect ratio and output quality. Apparently, despite all the hype, these models still struggle with basic instructions, proving that ‘improved’ doesn’t always mean ‘actually useful.’ <a href="https://app.daily.dev/posts/zRxAENzbc" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Developers fear AI will steal their jobs, but it’s just ‘Deep Blue’ anxiety</strong>: Software developers are apparently experiencing ‘Deep Blue’ anxiety, fearing AI coding agents like Claude Opus 4.5/4.6 and GPT-5.2/5.3 will devalue their skills. The article reassures them that accumulated experience is still valuable, even as AI handles routine coding. So, don’t worry, your job isn’t gone, it’s just… different now. You’re still special, just maybe not for writing boilerplate. <a href="https://app.daily.dev/posts/DwIZWcSFk" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Over-reliance on AI coding assistants will make you dumb</strong>: An opinion piece warns that using AI coding assistants too much will lead to ‘cognitive atrophy’ in programmers, making them lose problem-solving skills. The author compares it to GPS users losing their ability to navigate without a map. So, if you’re using Copilot, prepare to become a coding zombie, unable to learn new concepts. Because apparently, convenience is the enemy of intelligence. <a href="https://app.daily.dev/posts/qtVi65Njq" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
