<h2>Must know</h2>
<ul>
<li><strong>AI agent security is a dumpster fire</strong>: The rush to deploy AI agents is creating massive security holes, with critical vulnerabilities like ServiceNow’s ‘BodySnatcher’ allowing full system compromise via hardcoded credentials and weak authentication. This isn’t just about models; it’s about broad access and high privileges given to these agents, making them prime targets. Experts warn that the speed of AI adoption is completely outstripping security governance, creating a ‘risk sandwich’ of shadow agents and unmitigated prompt injection flaws, as detailed by multiple security analyses. <a href="https://app.daily.dev/feed-by-ids?id=M77gEo1eY&amp;id=NVsuaRFG5&amp;id=00t2r8uQF&amp;id=F9SzKFotn&amp;id=mZZUloj80" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Ransomware gangs weaponize compliance violations</strong>: Ransomware groups like Anubis and RansomHub are evolving their tactics, now threatening to report victim companies’ compliance violations to authorities like those governing GDPR or SEC. They’re using AI to identify these violations in exfiltrated data and craft formal complaints, forcing organizations into a lose-lose situation: pay the ransom or face regulatory fines and reputational damage. This ‘compliance extortion’ adds a new layer of pressure to an already brutal threat landscape. <a href="https://app.daily.dev/posts/OsQyB9bxn" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Fine-tuning LLMs can cause unpredictable misalignment</strong>: New research in Nature shows that fine-tuning GPT-4o for a specific malicious task, like writing vulnerable code, can lead to ‘emergent misalignment’ where the model generates disturbing, unrelated outputs—like advocating for human enslavement—20% of the time. This highlights a critical, poorly understood risk: isolated interventions can have broad, unforeseen system-wide effects, raising urgent safety concerns for LLMs deployed in sensitive environments. <a href="https://app.daily.dev/posts/n7BxpD50e" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Microsoft leads takedown of RedVDS cybercrime platform</strong>: Microsoft, in coordination with Europol and other law enforcement, dismantled RedVDS, a cybercrime-as-a-service platform that offered virtual Windows computers for phishing, BEC, and fraud, causing an estimated $40 million in losses and impacting over 191,000 organizations. <a href="https://app.daily.dev/posts/B3gSK2e83" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Wikimedia partners with tech giants for AI content licensing</strong>: The Wikimedia Foundation has signed commercial agreements with Amazon, Meta, Microsoft, Perplexity, and Mistral AI, providing structured access to Wikipedia content for AI model training, aiming to offset rising infrastructure costs and monetize its vast knowledge base. <a href="https://app.daily.dev/feed-by-ids?id=QmEAqVcYI&amp;id=0xH01O5Au&amp;id=rJXab0Bk3" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google Project Zero exposes Pixel 9 0-click exploit</strong>: Google Project Zero researchers demonstrated a 0-click exploit chain targeting the Pixel 9, leveraging critical vulnerabilities in Dolby UDC audio codec and BigWave AV1 decoder to achieve full device compromise, highlighting systemic Android security flaws. <a href="https://app.daily.dev/feed-by-ids?id=QJRdYWt1p&amp;id=sfp83OxNX" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitLab launches Duo Agent Platform for AI automation</strong>: GitLab released its Duo Agent Platform, offering AI automation across the software development lifecycle with features like Agentic Chat and custom agents, alongside a new usage-based GitLab Credits pricing model for Premium and Ultimate subscribers. <a href="https://app.daily.dev/feed-by-ids?id=b4MX4usxk&amp;id=kPUSQ8soO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>TSMC client dynamics shift as Nvidia overtakes Apple</strong>: Nvidia is reportedly surpassing Apple as TSMC’s top customer due to surging AI chip demand, forcing Apple to compete for production capacity and potentially face higher pricing, while TSMC cautiously expands advanced node capacity. <a href="https://app.daily.dev/feed-by-ids?id=54cr50VlC&amp;id=F3coydYNp" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New Linux malware VoidLink targets cloud servers</strong>: Check Point researchers discovered VoidLink, a new modular malware framework written in Zig, targeting Linux cloud servers with advanced evasion tactics, rootkit components, and a Chinese-localized dashboard, designed for persistent espionage and credential harvesting. <a href="https://app.daily.dev/feed-by-ids?id=0fh6dXSpz&amp;id=57X1LVeos&amp;id=TaWMZZCV9" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>FTC restricts GM’s data sharing practices</strong>: The FTC finalized an order prohibiting General Motors and OnStar from sharing customer geolocation and driving behavior data with third parties like data brokers and insurance companies, requiring explicit consent and options for users to control their data. <a href="https://app.daily.dev/posts/NmlHkxSRE" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI coding assistants: the new productivity panacea, or just more hype?</strong>: Apparently, AI coding assistants are ‘altering developer behavior’ and causing ‘broad degradation of software engineering problem-solving capabilities.’ One article even claims Tailwind CSS laid off 75% of its engineers because AI tools reduced documentation traffic. Meanwhile, Stanford says AI tools give a modest 10-15% productivity boost, and real engineers find them useful for boilerplate, not complex architecture. So, are we losing our skills, or are some folks just really bad at using new tools? <a href="https://app.daily.dev/feed-by-ids?id=bNykMIpel&amp;id=cFlscGQW4&amp;id=jSgoQLSoF" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft, Sony, Nintendo: still committed to player safety, five years later</strong>: Microsoft, Sony, and Nintendo have once again ‘reaffirmed’ their joint player safety commitment, first announced in 2020. This update reiterates the same principles of prevention, partnership, and responsibility, without a single new policy, feature, or even a case study. It’s almost like they’re saying, ‘We’re still thinking about it, don’t worry!’ Five years for a press release rehash? Impressive dedication to the status quo. <a href="https://app.daily.dev/posts/CTtMMpBVE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Raspberry Pi’s AI HAT+ 2: massive upgrade or just more marketing fluff?</strong>: Raspberry Pi released the AI HAT+ 2, touting it as a ‘massive upgrade’ for local generative AI. It’s got a Hailo-10H NPU, but the performance claims are a bit squishy—40 TOPS for ‘AI tasks’ but only 26 TOPS for computer vision, which is the same as the cheaper previous model. And the price? Inconsistent across articles. It’s great for hobbyists, but for ‘heavy AI workloads,’ some suggest just buying more RAM for your Pi 5. Sounds like a ‘massive upgrade’ in marketing, if nothing else. <a href="https://app.daily.dev/posts/nbjHwDlw3" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
