
amla-sandbox
WASM sandbox with capability enforcement for AI agent code. Agents can only call explicitly provided tools with defined constraints. Sandboxed…
ai-securitycontainer-securitygeneral-purpose-utilities+2
343

WASM sandbox with capability enforcement for AI agent code. Agents can only call explicitly provided tools with defined constraints. Sandboxed…

Run untrusted AI code safely, fast

Proof-of-concept demonstrating a Node.js permission model bypass (CVE-2026-21636) that allows network access via undici/fetch to local services,…

Let your AI go full send. Your home directory stays home.


Executes untrusted Python, JavaScript, and Bash in a hardened jail with 5-second, 128 MB, 0.5-core, PID and output limits.

CVE-2026-99999 Working PoC — real VM escape for Linux kernel csum_start bug (CVSS 9.3). Guest-to-Host heap leak. See also: CVE-2026-99999-analysis…
