<h2>Must know</h2>
<ul>
<li><strong>Pentagon-Anthropic AI use dispute escalates</strong>: The Pentagon is threatening to designate Anthropic as a supply chain risk because the company refuses to allow unrestricted military use of its Claude AI, specifically for mass surveillance of Americans or autonomous weapons without human oversight. This dispute, following Claude’s alleged use in a Venezuelan military operation, sets a precedent for AI companies attempting to limit government use of their technology, unlike OpenAI, Google, and XAI, which have agreed to ‘all lawful purposes’ terms. <a href="https://app.daily.dev/posts/sX4nQIjj6" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>European Parliament bans AI tools on devices</strong>: The European Parliament has disabled AI features like writing assistants and summarization tools on lawmakers’ corporate devices due to data security and privacy concerns. Their IT department cannot guarantee data security when AI assistants use cloud services, especially given that many AI company servers are US-based and subject to government data demands, highlighting broader European unease about relying on US tech companies for sensitive legislative data. <a href="https://app.daily.dev/posts/NW6nIRqB9" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agent security vulnerabilities and exploitation</strong>: New research and CTF challenges highlight critical security flaws in AI agents. Check Point Research demonstrated how AI assistants like Grok and Microsoft Copilot can be exploited as covert Command-and-Control proxies without API keys, tunneling malware communications. Meanwhile, a ‘Prompt Injection CTF’ is actively challenging participants to trick an AI assistant into revealing sensitive <code>secrets.env</code> file contents, underscoring the immediate threat of prompt injection and AI-driven malware. <a href="https://app.daily.dev/feed-by-ids?id=DCxMGa8ig&amp;id=H92K14NPx" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Anthropic releases Claude Sonnet 4.6</strong>: Anthropic released Claude Sonnet 4.6, an upgrade to Sonnet 4.5, offering near-Opus 4.6 intelligence at a lower cost, with enhanced coding, agentic workflows, and a 1M token context window in beta. <a href="https://app.daily.dev/posts/TzfkywkQu" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Alibaba releases Qwen3.5 multimodal AI model</strong>: Alibaba launched Qwen3.5, a new multimodal AI model for enterprise agent workflows, claiming benchmark improvements over GPT-5.2, Claude 4.5 Opus, and Gemini 3 Pro, with expanded language support and a 1 million token context window. <a href="https://app.daily.dev/posts/UOocLbfK4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI’s impact on agile software development debated</strong>: While some argue AI agents building applications in hours make the Agile Manifesto obsolete, others propose ‘augmented coding’ and ‘Intent Design’ as evolutions, with Forrester reporting 95% of professionals still find Agile relevant and nearly half integrating generative AI. <a href="https://app.daily.dev/posts/X6nRpouih" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Heroku launches AI agent code execution sandboxes</strong>: Heroku introduced code execution sandboxes for AI agents to securely run untrusted code in isolated dynos, aiming to reduce token consumption by up to 98% by allowing agents to orchestrate operations via scripts rather than passing all data through the model’s context window. <a href="https://app.daily.dev/posts/bzMzHxxko" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitHub enhances enterprise security and governance</strong>: GitHub rolled out new features for Enterprise Cloud, including a required reviewer rule for granular approval policies, enterprise-wide credential management for security incidents, and improved custom properties and rule insights for better organizational policy enforcement. <a href="https://app.daily.dev/feed-by-ids?id=GdbnCMd5H&amp;id=O4y8yihCh&amp;id=xF6kLcdZX" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>JDK 27 integrates post-quantum hybrid key exchange</strong>: JDK 27 introduces JEP 527, integrating post-quantum hybrid key exchange into TLS 1.3 to combine quantum-resistant ML-KEM with traditional elliptic-curve algorithms, protecting against future quantum computing attacks like ‘harvest now, decrypt later’ without requiring application code changes. <a href="https://app.daily.dev/posts/duMdZgpLR" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OWASP smart contract top 10 2026 released</strong>: OWASP released its Smart Contract Top 10 2026, identifying governance failures, privilege misconfiguration, and upgrade authority concentration as dominant security risks in blockchain systems, based on hundreds of millions in losses from 2025 exploits. <a href="https://app.daily.dev/posts/U7Z7c8wKd" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI will automate everyone else’s job, not mine</strong>: The ‘GeLLMan Amnesia’ effect suggests people believe AI will automate others’ jobs but not their own, because AI’s flaws are obvious in their deeply known fields. It’s almost as if experts understand the nuances of their work better than a marketing department trying to sell you on the next big thing. Shocking, I know. <a href="https://app.daily.dev/feed-by-ids?id=7hHra6Ai0&amp;id=xjDRiBzgJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-resistant tech job identified</strong>: An article titled ‘The Only Tech Job That Survives AI’ claims architecture and system design skills are AI-resistant because they involve integrating platforms, evaluating vendors, and negotiating contracts. Because, clearly, AI can write code but can’t possibly handle a PowerPoint presentation or a vendor lunch. Good to know our jobs are safe as long as we keep talking about ‘synergy’ and ‘leveraging best practices.’ <a href="https://app.daily.dev/posts/smOggDoX8" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI coding assistant utility and perceptions</strong>: AI coding assistants are widely criticized as ‘overhyped’ in 2026, yet the author ‘secretly’ uses them for boilerplate, debugging, and automating release workflows. It’s almost like these tools are useful for the grunt work nobody wants to do, but admitting it would ruin the narrative of AI being either a magical genie or a job-stealing demon. Can’t we just call them ‘fancy autocomplete’ and move on? <a href="https://app.daily.dev/feed-by-ids?id=CBxVjaadg&amp;id=urOGgKvla" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
