| |
|
|
| |
| |
| CloudSecList is a newsletter for busy professionals who want to keep up to date with cloud security and its intersection with AI. Hand-curated by Marco Lancini since 2019. |
| |
| |
| Sponsor Important prereq for deploying AI: have a unified identity layer You’re under pressure to deploy autonomous agentic AI within your enterprise infrastructure environment ASAP. They’re asking at every meeting: what do you need to make it happen? Show them the Agentic Identity Framework from Teleport: Standards-based, opinionated designs that let you ship confidently without reinventing security. It all starts with a unified identity layer. Explore the Framework |
| |
| This week's articles Building an evidence-grounded agentic security operations harness on Cloudflare Cloudflare Managed Defense uses a team of specialized AI agents built on Workers and global network telemetry to analyze security alerts. By separating deterministic evidence collection from model inference, the system delivers grounded recommendations to analysts. The AI SDLC transformation playbook Atlassian's shared their own version of an AI SDLC playbook, which outlines transforming the full dev lifecycle using agentic AI, requiring two platform investments (context graph, measurement) and workflow shifts across planning, design, development, review, and maintenance, with humans retaining governance and judgment. Agent containment best practices Private beta guidance from Anthropic: run autonomous agents in hardware-virtualized sandboxes, one microVM per agent; keep credentials out of agent environments; use auto mode as extra check; retain transcripts 30+ days; report out-of-scope behavior. Designing MCP Gateway Uber's MCP Management Platform Uber built MCP Gateway, a centralized microservice hosting 800+ MCP servers and 5000+ tools. It auto-discovers internal APIs via IDL crawling, translates HTTP/gRPC/TChannel calls to MCP, enforces auth/redaction, and provides runtime discovery via Omni MCP to reduce agent context bloat. Pretty Themes, Hidden Loaders: GlassWorm-Linked Extensions Span VS Code Marketplace and Open VSX Socket uncovered a GlassWorm-linked cluster of malicious VS Code themes spanning VS Code Marketplace and Open VSX. Extensions like Coca-Cola Christmas and Cosmic Nebula Themes embed obfuscated loaders using AES-256-CBC decryption, Solana dead-drop infrastructure, and Windows command execution to deliver remote payloads. Azure's Weakest Link - Five Full Cross-Tenant Compromises Five critical Azure cross-tenant privilege escalation vulnerabilities in API Connections (Logic Apps) were found via path traversal in DynamicInvoke, undocumented Dynamic* endpoints, and the host.api.RuntimeUrl parameter, yielding $200,000 in bug bounties. The world hasn’t figured out yet that you can literally just fix everything with a Nix overlay Nix overlays let you patch any software at any level, from userland tools to OpenSSL, across devenv, Docker images, and NixOS VMs, all from a single source of truth. To be fair, AI makes Nix's steep learning curve far more approachable now. |
| |
| |
| Sponsor Stop reinventing your first year of security Companies at the same stage need the same foundations: SSO, offboarding, patch cadence, logging, incident response. You still have to work out the order and write the plan from scratch. PrimeOutpost ships that first year as a sequenced roadmap with playbooks, fitted to your industry, headcount and stack. We are taking on a few design partners: full product free, plus a 30-minute onboarding call, in return for honest feedback. Apply as a design partner |
| |
| Tools VulnHunter Agentic AI security scanner that hunts exploitable vulnerabilities like an adversary, proves them with executable PoCs, and fixes them test-first. quicksand Quickly sandbox your AI agent. Zurp Meta bug bounty research tools, in Burp Suite and in your coding agent. You can also check out the companion blog post. EntraGoat A Deliberately Vulnerable Entra ID Environment. |
| |
| From the cloud providers #AWS Introducing Strands Box: AI agent sandboxes powered by Dogwood Strands Box is an open source (Apache 2.0) developer preview sandbox for AI agents. It combines OS-level isolation (such as macOS Seatbelt) with Dogwood policies that govern actions at egress, Python, Shell, and MCP enforcement points. #AWS Building your AI vulnerability harness The article describes a three-layer pipeline that narrows scanner findings through multi-scanner agreement, AST-based structural verification against code, and deployment context from IaC controls such as WAF rules. It outputs a small, prioritized set of evidenced findings, which still needs human review. You can also check out Part 2. #AZURE 3 lessons from frontier AI vulnerability research FORGE Lab reports 140 Windows CVEs and 155 validated open-source reports. Three lessons: measure the full pipeline, not just findings; spend reasoning only where it removes uncertainty; and make validation and remediation a continuous learning loop. |
| |
| Thanks for reading! |
| If you found this newsletter helpful, I'd really appreciate if you could forward it to your friends and colleagues! 👌 If you have questions, comments, or feedback, let me know on Twitter (@lancinimarco / @CloudSecList), or at feedback.cloudseclist.com! Thanks, Marco |
| |
| |
| |
| | | |
| |
| |
|
| Forward |
| |
| |
|
| Tweet |
| |
| |
|
| Share |
| |
| |
| How did you like this issue of CloudSecList? 1 2 3 4 5 |
| |
| Unsubscribe from CloudSecList |
| Archives | View in browser | Sponsorship |
| © 2019-present, CloudSecList by Marco Lancini. |