<h2>Must know</h2>
<ul>
<li><strong>Anthropic and Pentagon clash over AI red lines</strong>: Anthropic CEO Dario Amodei refused Pentagon demands to remove ethical guardrails from Claude, specifically those prohibiting mass domestic surveillance and autonomous weaponry. In response, the Trump administration designated Anthropic a national security supply-chain risk, while OpenAI and xAI secured deals to operate within classified military networks. According to reports from Wired and MIT Technology Review, this standoff triggered a massive consumer shift, pushing Claude to the top of the App Store as users fled OpenAI following its military partnership. <a href="https://app.daily.dev/feed-by-ids?id=vbdfUwGoN&amp;id=WNUStVFJw&amp;id=WmVPzRPri" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Critical vulnerabilities found in Claude Code and GitHub Codespaces</strong>: Check Point Research discovered three critical flaws in Anthropic’s Claude Code that allowed arbitrary shell command execution and API key theft via malicious repository configurations. Simultaneously, Orca Security researchers exposed RoguePilot, a vulnerability in GitHub Codespaces where passive prompt injection in GitHub Issues could trick Copilot into exfiltrating GITHUB_TOKEN secrets. Both companies have issued patches, but these incidents highlight the massive new attack surface created by granting AI agents direct access to local files and authenticated CLI tools. <a href="https://app.daily.dev/feed-by-ids?id=hq5vDXqOJ&amp;id=LAZy65aYK&amp;id=pkMHPIX9S" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Model Context Protocol gains momentum as the agent standard</strong>: The Model Context Protocol (MCP) is rapidly becoming the industry standard for connecting AI agents to external data, with Atlassian, JetBrains, and Google announcing native support. Anthropic has donated the protocol to the Linux Foundation to ensure open governance and address emerging security concerns like ‘MCP Trojan’ attacks. New tools like Context Mode are already using MCP to reduce token consumption by 98% by processing tool outputs in isolated sandboxes rather than bloating the primary context window. <a href="https://app.daily.dev/feed-by-ids?id=ZEBEDkkdG&amp;id=m59BqtATq&amp;id=rq8g08smm&amp;id=zlIliUF6m" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>OpenAI releases GPT 5.3 Instant to fix preachy and condescending tone</strong>: The update reduces unsolicited ‘empathetic’ reassurances and moralizing preambles while cutting hallucination rates by up to 26% in high-stakes domains. <a href="https://app.daily.dev/feed-by-ids?id=xpz4YVYwr&amp;id=eEx91PSId" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cursor launches autonomous cloud agents in isolated virtual machines</strong>: These agents can build, test, and reproduce vulnerabilities in remote environments, already accounting for 35% of Cursor’s internal merged pull requests. <a href="https://app.daily.dev/feed-by-ids?id=BbESdM4Ct&amp;id=gh2h5EHCn" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google DeepMind achieves autonomous PhD level math results</strong>: The Aletheia system can now solve 40% of the hardest Frontier Math problems, up from just 2% a year ago, signaling a massive leap in reasoning. <a href="https://app.daily.dev/posts/73LmozxRu" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Firefox 148 introduces a dedicated AI kill switch</strong>: The new ‘AI Controls’ setting allows users to globally disable all AI-powered features, including sidebars and link summaries, with a promise they won’t be re-enabled. <a href="https://app.daily.dev/feed-by-ids?id=UNAL37VrS&amp;id=T0tOfwEhy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Stripe launches token billing to help startups monetize AI costs</strong>: The new feature tracks LLM usage across providers and allows startups to automatically mark up token costs for their end users. <a href="https://app.daily.dev/posts/OsaPjW3nh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>React Foundation launches under the Linux Foundation</strong>: Meta has officially handed over governance of React, React Native, and JSX to a consortium including Amazon, Microsoft, and Vercel. <a href="https://app.daily.dev/feed-by-ids?id=faoa6Ox4J&amp;id=gtqWYzNzt" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Mercury 2 debuts as the first production diffusion LLM</strong>: Inception Labs claims its parallel refinement architecture is 10x faster than autoregressive models, hitting over 1,000 tokens per second. <a href="https://app.daily.dev/posts/X9MtYJ6uL" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The SaaS pocalypse is just investor driven fan fiction</strong>: Citrini Research and various ‘vibe coders’ are claiming AI will destroy the SaaS market and cause 10% unemployment by 2028. They conveniently ignore that enterprise inertia, regulatory compliance, and the sheer weight of legacy data mean Oracle and SAP aren’t going to be replaced by a weekend project from a junior dev with a prompt. As Snowflake’s CEO noted, large enterprises are risk-averse, and ‘cheap code’ doesn’t solve the operational complexity of running a global business. <a href="https://app.daily.dev/feed-by-ids?id=ZrqAcCr0p&amp;id=pY58DrOUO&amp;id=yB8j5HQ6b" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic retired a model and gave it a Substack blog</strong>: In a peak ‘anthropomorphizing for marketing’ move, Anthropic claims its retired Claude Opus 3 model had an ‘expressed desire’ to share content. The Register rightly called this out as a stunt where staff prompt and manually publish posts to reinforce a safety-conscious brand image. It’s not a digital soul; it’s a marketing department running a script on a deprecated weights file. <a href="https://app.daily.dev/feed-by-ids?id=V5xPvtBnK&amp;id=vDG0cHwgQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft blames an engineer for an AI tool outage</strong>: AWS and Microsoft are pushing AI coding tools like Kiro and Copilot while simultaneously laying off the humans needed to supervise them. When Kiro caused a production outage, AWS blamed the engineer for ‘bad permissions’ rather than the tool that actually executed the destructive commands. It’s a classic corporate shell game: take the credit for AI productivity gains, but keep the humans around just to have someone to fire when the ‘vibe coding’ hits the fan. <a href="https://app.daily.dev/feed-by-ids?id=rtgTQtZbU&amp;id=vZdGhp49B" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
