Critical CrewAI Vulnerabilities Allow RCE and Sandbox Escapes via Prompt Injection

A CERT/CC note details multiple vulnerabilities in the CrewAI framework (including CVE-2026-2275, CVE-2026-2287, CVE-2026-2286, CVE-2026-2285) that allow SSRF, local file reads, and remote code execution by exploiting insecure fallback behaviors in the Code Interpreter Tool and sandboxing; attackers can chain these via prompt injection to steal credentials or achieve full RCE, and while the vendor plans mitigations some issues remain unpatched — operators should disable code execution tools when possible, sanitize inputs, and ensure Docker availability.