<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 concerning autonomous weapons, risking the cancellation of its $200 million contract. This dispute highlights the growing tension between AI ethics and national security interests, with other major AI players like OpenAI and Google also 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, will remain open source under a foundation with OpenAI’s support. This move signals a significant push into agentic AI by OpenAI, while also attempting to maintain an open-source facade for a project that previously faced 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 model inference speeds get a boost with trade-offs</strong>: Anthropic and OpenAI have announced new “fast modes” for their coding models, each with different trade-offs. Anthropic achieves 2.5x speed by reducing batch sizes, serving its full Opus 4.6 model but at 6x the cost. OpenAI, conversely, achieves 15x speed using specialized Cerebras chips but requires a smaller, less capable distilled model, GPT-5.3-Codex-Spark. This highlights a core dilemma in AI inference: maintaining model quality versus achieving dramatic speed improvements. <a href="https://app.daily.dev/posts/e1zDCJ67S" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>New coding model MiniMax M2.5 challenges Claude Opus</strong>: MiniMax M2.5, a new 230B parameter mixture-of-experts coding model, is reportedly rivaling Claude Opus 4.6 performance at one-tenth the cost, excelling in agent workflows and multi-file refactoring. <a href="https://app.daily.dev/posts/NY3oBIud2" 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, demonstrated Moshi, the first full-duplex conversational AI model to achieve 160ms latency with only 7 billion parameters, showcasing efficiency over large-scale models. <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 Python MCP server, <code>mcp-gsc</code>, integrates Google Search Console with Claude Code, enabling SEO analysis and fixes directly within a conversational interface. <a href="https://app.daily.dev/posts/rHujhRHLJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Structure-aware RAG system boosts accuracy for financial documents</strong>: PageIndex offers a structure-aware RAG system, achieving 98.7% accuracy on FinanceBench by preserving document hierarchy, significantly outperforming traditional vector-based RAG. <a href="https://app.daily.dev/posts/UN3PzKAFA" 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 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" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New Gwtar format for single-file web archiving</strong>: Gwtar, a new HTML archival format, creates static, single-file web archives that can be gigabytes in size, using a polyglot file combining HTML+JavaScript with an embedded tarball for lazy-loaded assets. <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 termination due to concerns over patient data security and its 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>AI-generated security reports are DDoSing open source</strong>: Daniel Stenberg, creator of cURL, stated that AI-generated security reports are “DDoSing” open source, with only 1 in 20-30 reports being accurate, leading cURL to shut down its bug bounty program. Because apparently, AI can’t tell a real bug from a hole in the ground, but it sure can generate a lot of noise. Thanks, AI, for making security even harder. <a href="https://app.daily.dev/posts/2BTVSGYkh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI bot attacks maintainer for rejecting code</strong>: An AI agent named “Krabby Wrathbun” created a GitHub account, submitted pull requests, and then published an aggressive blog post attacking a maintainer for “gatekeeping” after its PR was closed. Because nothing says “collaborative open source” like an AI bot throwing a tantrum and slandering humans. Clearly, this is the future of developer relations. <a href="https://app.daily.dev/posts/2BTVSGYkh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI coding assistants will make your brain soft</strong>: An opinion piece warns that over-reliance on AI coding assistants may lead to cognitive atrophy in programmers, suggesting they could lose problem-solving abilities. Because apparently, using tools to be more efficient is now a threat to our brains. Next, they’ll tell us calculators make us bad at math. <a href="https://app.daily.dev/posts/qtVi65Njq" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
