<h2>Must know</h2>
<ul>
<li><strong>AI agent security flaws and emerging threats</strong>: Security researchers from Wiz breached an AI agent social network, Moltbook, in under three minutes by exploiting basic security design flaws, demonstrating how AI coding assistants can be used for both development and exploitation. This incident, along with the emergence of “prompt worms”—self-replicating prompts targeting AI agents’ instruction-following—highlights critical and evolving security challenges for AI agent platforms, as reported by Wiz. Meanwhile, Gravitee research indicates 1.5 million of 3 million AI agents in large enterprises operate without oversight, with 88% experiencing security incidents. <a href="https://app.daily.dev/feed-by-ids?id=ycPDVy6OP&amp;id=Hv04ODVQw&amp;id=ehksC8VMf&amp;id=gy7WeHs5A&amp;id=kssiMjacP" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Deno Deploy reaches GA, launches sandbox for untrusted code</strong>: Deno Deploy is now generally available, offering zero-config deployment for JavaScript and TypeScript applications with Node.js framework support and integrated serverless PostgreSQL. Crucially, Deno Sandbox was introduced in beta, providing instant Linux microVMs for securely running untrusted code, particularly LLM-generated code, with hypervisor-level isolation and network egress control. <a href="https://app.daily.dev/posts/Jmdw76tdu" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Databricks Lakebase and Snowflake expand AI data cloud offerings</strong>: Databricks Lakebase is now generally available on AWS, offering a serverless Postgres service that scales automatically and runs directly on data lakes, unifying operational, analytical, and AI workloads. Similarly, Snowflake launched Cortex AI, a serverless generative AI suite, and a PostgreSQL database-as-a-service within its AI Data Cloud, both designed to enable AI applications and data analysis without external data movement, reducing latency and compliance risks. <a href="https://app.daily.dev/feed-by-ids?id=qtDOznoeU&amp;id=1F9HXgJiK&amp;id=ppuh0CVeq&amp;id=qgdGrzKKG" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>OpenAI details Codex AI agent internal design</strong>: OpenAI published the first article in a series explaining the internal design of Codex, their software development agent, detailing its agent loop architecture and strategies for managing LLM inference performance. <a href="https://app.daily.dev/feed-by-ids?id=DWHoaTisZ&amp;id=Dick5lxkX&amp;id=JSYEwKNu7&amp;id=YQpwOBE0D&amp;id=aSHfIm3oS&amp;id=cBoJTA7Xz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vercel v0 platform revamped for enterprise AI development</strong>: Vercel rebuilt its v0 AI-powered development platform, evolving it from a prototyping tool to an enterprise-grade AI coding platform with security features, integrations with Snowflake and AWS, and Git-based workflows for production-ready code. <a href="https://app.daily.dev/posts/8wRm1h8NM" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AWS IAM Identity Center adds multi-region replication</strong>: AWS IAM Identity Center now supports multi-Region replication for workforce identities and permission sets, enhancing resiliency for AWS account access and helping meet data residency requirements without additional cost. <a href="https://app.daily.dev/posts/wVGc4PkCH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OIDC adoption for machine identities and cloud security</strong>: Dependabot now supports OpenID Connect (OIDC) authentication for private registries, allowing dynamic acquisition of short-lived credentials and reducing the attack surface, while Bitbucket Pipelines also added support for multiple audiences in OIDC claims to simplify authentication across services. <a href="https://app.daily.dev/feed-by-ids?id=P7dsW78XN&amp;id=YmYpo2Y1G&amp;id=iDGQonMWb" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Critical IT vulnerabilities update</strong>: Cyble tracked 1,147 vulnerabilities last week, with 128 having public proof-of-concept exploits, including critical flaws in n8n, WinRAR, and OpenSSL, while CISA added several to its Known Exploited Vulnerabilities catalog. <a href="https://app.daily.dev/posts/OEq44lx9O" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Ubisoft union calls for CEO resignation</strong>: Union representatives at Ubisoft Paris are calling for CEO Yves Guillemot to resign due to a “complete breakdown of trust” stemming from studio closures, game cancellations, a five-day return-to-office mandate, and allegations of stifling creativity. <a href="https://app.daily.dev/posts/Ko8v1dV9o" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GNU Hurd project shows significant progress</strong>: The GNU Hurd project, a microkernel-based Unix replacement, is showing significant progress with 80% Debian package compilation, support for major desktop environments, and completed x86_64 and SMP functionality, countering perceptions of stagnation. <a href="https://app.daily.dev/posts/XjM52iBH8" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>X replaces search with Grok LLM</strong>: X (formerly Twitter) has apparently replaced its entire search functionality with Grok, an LLM, because fixing the existing search system was “too difficult.” So, instead of engineering a robust search, they just threw billions at an LLM and called it a day. What could possibly go wrong with a hallucinating search engine? <a href="https://app.daily.dev/posts/1TELuPW5m" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Take-Two Interactive addresses generative AI in game development</strong>: Take-Two Interactive CEO Strauss Zelnick stated that generative AI plays “zero part” in GTA 6 development, emphasizing its “entirely handcrafted” world, while simultaneously admitting they use general AI tools for “cost and time efficiencies” across hundreds of pilot programs. So, AI is useless for creativity but great for cutting costs. Got it, just like every other tech company trying to sound innovative without actually innovating. <a href="https://app.daily.dev/posts/chwMGhSO9" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Polish railway portal randomizes schedules for mobile users</strong>: A Polish railway passenger portal is displaying randomized train schedules to mobile users from specific IP addresses, with departure times varying by ±60 minutes. PKP declined to comment, citing “security reasons.” Apparently, confusing your customers with random data is now a valid cybersecurity strategy to combat scraping. Peak user-centric design, folks. <a href="https://app.daily.dev/feed-by-ids?id=NpQixwGEK&amp;id=hQA36mWZO" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
