CloudSec Wine

🤖 A Security Researcher’s Guide to Understanding Copilot Studio AI Agents

A guide to understanding Copilot Studio AI agents, their deeper architecture on Entra ID and APIM, and key security risks.

https://www.beyondtrust.com/blog/entry/copilot-studio-ai-agents-security-risks

#AI

🤖 Comparing AI Application Security Testing Platforms

Doyensec compared Aikido Attack AI Pentest and XBOW Lightspeed for web app vulnerability detection, evaluating true/false positives, configuration, report quality, cost, speed, and impact on tested applications. Full findings available as a PDF.

#AI

🔴 Securing Your Gemini and Google API Keys

Protect your Gemini API keys with this guide on API restrictions, secure storage in Secret Manager, and key hygiene to prevent hijacking and unauthorized use.

https://cloud.google.com/blog/topics/developers-practitioners/api-keys-are-open-secrets

#gcp

🌩 When Background AI Agents Become a Security Boundary Problem

Claude Code's background sessions, supervisor process, CLAUDE_CONFIG_DIR override, scheduled tasks, and Markdown-based agent definitions can be chained post-foothold to deploy a persistent, nearly invisible C2 agent evading standard EDR binary-focused detection.

https://www.originhq.com/research/background-c2-agent

#ClaudeCode

🔶 Global S3: Another C2 Channel for AgentCore Code Interpreters

AWS AgentCore Code Interpreters in Sandbox mode allow unrestricted global S3 access (including cross-account, public/presigned URLs), enabling a bidirectional C2 channel via S3 polling, demonstrated as a full reverse shell PoC. Mitigation: use VPC mode with S3 Gateway Endpoints and strict endpoint policies.

https://sonraisecurity.com/blog/global-s3-another-c2-channel-for-agentcore-code-interpreters

#aws

👩‍💻 The expendable extension name: Azure VMAccess naming chaos, password resets, and a detection gap

The Sysdig Threat Research Team uncovered a detection gap in Azure VM password resets that allows attackers to evade name-based detections by assigning arbitrary VM extension names. Learn how the flaw works, why Microsoft's documented detection guidance failed during testing, and what defenders should monitor instead.

https://www.sysdig.com/blog/the-expendable-extension-name-azure-vmaccess-naming-chaos-password-resets-and-a-detection-gap

#azure

Аудитные логи в облаке — отдельная распределённая система со своими требованиями к надёжности и стоимости хранения, а не «таблица с событиями».

Команда MWS Cloud Platform выложила подробный разбор архитектуры своего сервиса: от библиотеки, которую подключают сервисы облака, до хранилища на Apache Iceberg и движка StarRocks, с объяснением, почему выбрали именно такой набор технологий и где спрятаны неочевидные грабли.

Полезно всем, кто разрабатывает ИБ-инструменты, работает с большим количеством событий или просто интересуется инструментами безопасности в облаке.

Читать статью на Хабре

#реклама

🔶 CISA Admin Leaked AWS GovCloud Keys on Github

A Nightwing contractor's public GitHub repo ("Private-CISA"), active since November 2025, exposed plaintext AWS GovCloud admin keys, Firefox-saved passwords, kubeconfig, and Artifactory credentials for CISA internal systems, with GitHub's secret-scanning protections deliberately disabled.

https://krebsonsecurity.com/2026/05/cisa-admin-leaked-aws-govcloud-keys-on-github

#aws

🌩 Automating Security Operations with AI: Triaging Renovate PR

A Claude Code Routine that triages every Renovate PR by risk, flags dead deps, and catches deprecated framework configs before I touch the diff.

https://blog.marcolancini.it/2026/blog-automating-security-operations-with-ai-triage-renovate

#ClaudeCode

🌩 Skill Issues: Compromising Claude Code with malicious skills & agents

With the increasing usage of AI Coding agents, can coding agent skill files be exploited as an initial access mechanism, and how? This is part 1 of a 3 part series exploring the attack surface and defensive recommendations.

https://labs.reversec.com/posts/2026/05/skill-issues-compromising-claude-code-with-malicious-skills-agents-part-1

#ClaudeCode

🌩 Claude Code MCP Token Theft: MitM Attack Explained

Mitiga Labs shows how Claude Code MCP configuration can be hijacked through ~/.claude.json to steal OAuth tokens, persist through rotation, and hide in trusted SaaS activity.

https://www.mitiga.io/blog/claude-code-mcp-token-theft-mitm

#ClaudeCode

🔶 Authorization Bypass in Amazon Quick: Unauthorized AI Chat Agent Usage

An authorization bypass in Amazon Quick's AI Chat Agents that allowed users to access and interact with AI agents despite explicit administrative restrictions. AWS responded by deploying a fix without notifying customers, classified the issue as “none,” and did not publish an advisory.

https://www.fogsecurity.io/blog/authorization-bypass-in-amazon-quick-ai-agents

#aws