<h2>Must know</h2>
<ul>
<li><strong>Kubernetes telemetry exploit allows full cluster compromise</strong>: A security researcher uncovered an exploit in Kubernetes’ <code>nodes/proxy GET</code> call that enables privileged command execution and full cluster compromise without leaving audit trails. Despite the severe impact, the Kubernetes team labeled it “intended behavior,” not a bug, and no CVE was issued. A fix, KEP-2862, is planned for Kubernetes 1.36, but until then, administrators should audit RBAC policies and restrict <code>kubelet</code> port access to mitigate the risk, as reported by multiple security outlets. <a href="https://app.daily.dev/posts/bOiz0Chx6" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-powered polymorphic phishing attack emerges</strong>: A new polymorphic phishing attack leverages AI models like DeepSeek and Gemini to dynamically generate malicious JavaScript code directly within victims’ browsers. This technique bypasses traditional security measures by splitting malware into fragments delivered via separate AI prompts and reassembled client-side from trusted LLM domains, a process termed “last mile reassembly.” This highlights the increasing use of AI in offensive security operations, as detailed by various cybersecurity reports. <a href="https://app.daily.dev/posts/jfmiKNwaJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google BigQuery integrates SQL-native AI inference</strong>: Google BigQuery now supports SQL-native managed inference, allowing data teams to deploy and run over 180,000 Hugging Face and Vertex AI Model Garden models directly via SQL statements. This eliminates the need for separate ML infrastructure, as BigQuery automatically provisions and manages compute resources. This move positions BigQuery as a direct competitor to Snowflake Cortex AI and Databricks Model Serving, simplifying AI adoption within data workflows, according to Google’s announcement. <a href="https://app.daily.dev/posts/eM9NtN7mD" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>SERA open-source coding agent announced</strong>: SERA, an open-source coding agent, achieves state-of-the-art performance on SWE-Bench Verified at 54.2% and is affordable to train, costing approximately $400 for reproduction. <a href="https://app.daily.dev/posts/y5jjtMeVw" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Dynatrace launches AI observability app and agentic operations platform</strong>: Dynatrace released its AI Observability app and a new agentic operations platform, providing end-to-end visibility for agent interactions, tool usage, and cost tracking across major cloud providers, aiming to shift to autonomous operations. <a href="https://app.daily.dev/posts/d1KKTrRCA" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google launches agentic vision for Gemini 3 Flash</strong>: Google introduced Agentic Vision for Gemini 3 Flash, transforming static image processing into interactive investigation by combining visual reasoning with Python code execution for tasks like image annotation and diagram interpretation. <a href="https://app.daily.dev/posts/KIgS0LnaZ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Moonshot AI releases Kimi K2.5 visual agentic model</strong>: Moonshot AI launched Kimi K2.5, an open-source visual agentic model with a Mixture of Experts architecture and a 256K context length, featuring an Agent Swarm capability for coding, multimodal reasoning, and research automation. <a href="https://app.daily.dev/posts/2GP3EQ58p" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Enterprise AI agent rollouts face production challenges</strong>: Enterprise AI agent implementations are struggling to transition from lab to production, primarily due to challenges in aggregating unstructured data and replicating human judgment, leading to a “trough of disillusionment” regarding ROI. <a href="https://app.daily.dev/posts/7PYYjzg77" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microservices orchestration simulator</strong>: An interactive simulator demonstrates Kubernetes orchestration concepts like reconciliation loops, self-healing, horizontal pod autoscaling, and rolling updates, visualizing how declarative infrastructure management maintains desired state. <a href="https://app.daily.dev/posts/VIRS5PfQf" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Terraform challenger Formae expands to more clouds</strong>: Platform Engineering Labs expanded its Infrastructure as Code platform, Formae, to support Google Cloud, Azure, Oracle Cloud, and OVHcloud, offering an alternative to Terraform with improved state management through autonomous agents and AI-enhanced tooling. <a href="https://app.daily.dev/posts/qaAQ3pbgp" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Anthropic CEO advocates minimal AI regulation amid profitability concerns</strong>: Anthropic CEO Dario Amodei, whose company isn’t expected to turn a profit until 2028, just dropped a 22,000-word essay arguing for ‘minimal AI regulation.’ Apparently, the real threat is theoretical superintelligent AI, not, you know, actual problems like wealth inequality or job displacement. Critics are calling it a ‘thinly veiled screed against regulation,’ which sounds about right when your business model relies on unchecked growth, as noted by articles like #amOI7bITg. <a href="https://app.daily.dev/posts/amOI7bITg" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Moltbot AI assistant faces security and trademark challenges</strong>: Moltbot, the viral open-source AI assistant, is making headlines for all the wrong reasons: critical security vulnerabilities, plaintext storage of sensitive data, and its GitHub account getting hijacked by crypto scammers. Oh, and Anthropic forced a name change because it was too ‘Claude-inspired.’ So, we have a popular tool that’s a security nightmare, a trademark infringement, and a target for malware. What could possibly go wrong, as highlighted in articles like #zkO1JWwHu? <a href="https://app.daily.dev/posts/zkO1JWwHu" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI integration enhances non-human identity compliance: the magic bullet for security</strong>: Apparently, integrating AI into Non-Human Identity (NHI) management is the silver bullet for cybersecurity, promising predictive analytics, automated threat detection, and dynamic security responses. The article, conveniently lacking specific sources or authors, frames this as universally beneficial with no downsides. It’s almost as if ‘AI’ is the new ‘blockchain’ – just sprinkle it on anything, and suddenly it’s a ‘scalable solution’ that ‘adapts to evolving cyber threats,’ as vaguely described in #acGpetBEA. <a href="https://app.daily.dev/posts/acGpetBEA" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
