EGRET ARTICLES
Articles about AI & agents.
Clear findings. Sources you can check.
Google Threat Intelligence and Mandiant say attackers supplied compromised credentials to agentic pentesting frameworks such as ARTEX, letting the tools autonomously move through victim networks; one Wiz-tracked campaign reached…
Google Threat Intelligence
A local attacker can make the MCP filesystem server leak files outside its allowed directories: it checks a path, then reopens it by name, allowing a directory-to-symlink swap between the check and the open, according to advisory…
MCP server advisoriesThe address guard in mcp-server-fetch wrongly classifies an internal IPv6 site-local address as public, letting a fetch bypass the SSRF check meant to refuse non-public destinations, researchers at ShieldPi report.
MCP server advisoriesA read-only request for Git repository status can run a program chosen by the repository itself. The advisory reports that mcp-server-git lets an untrusted repository's own .git/config execute arbitrary code through…
MCP server advisoriesA local user on shared Linux or macOS hardware can read sensitive Aspire CLI configuration because the tool writes diagnostics to a fixed temporary-file path, Microsoft's advisory says. Microsoft attributes the flaw to the CLI's…
GitHub Advisory DatabaseA file-read tool in an AI code assistant can be tricked into reading any file on the server, not just files inside the project folder. GitHub's advisory database reports that the flaw requires no authentication when the HTTP…
GitHub Advisory DatabaseA single forged X-Forwarded-For header lets any remote caller defeat ezBookkeeping’s configured IP allowlist and have actions logged under a trusted address instead of the attacker’s own IP, according to the GitHub advisory.
GitHub Advisory DatabaseA remote Telegram user who can message praktor by default can write files outside the uploads directory into the agent's persistent workspace, privately reported by tonghuaroot, because the empty allow list permits every sender.
GitHub Advisory DatabaseThe MCP filesystem tool list_directory_with_sizes follows symlinks when listing children, so an inside symlink to an outside file leaks that file's size. Researchers @cybercraftsolutionsllc reproduced it on…
MCP server advisoriesA malicious prompt can make the banks LLM template library run arbitrary code through its tool-handling path, GitHub's advisory warns. The flaw survives the 2.4.2 Jinja sandbox patch because it resolves import_path in Python…
GitHub Advisory DatabaseOriginal advisories and research, awaiting Egret editorial review.