The SendEmailBlock accepts a user‑supplied SMTP server address and port and forwards them directly to Python's smtplib without performing any IP validation. This bypasses AutoGPT's hardened SSRF protections that normally block connections to private, loopback, link‑local and cloud metadata addresses. As a consequence, an authenticated user on a shared deployment can instruct the platform to open a raw TCP connection to any internal host, read the service banner on connect, and display the banner in the block output, effectively providing non‑blind internal port scanning and service fingerprinting.

The vulnerability affects Significant‑Gravitas AutoGPT platform versions 0.1.0 through 0.6.51. It is fixed in version 0.6.52 and later.

With a CVSS score of 5 the issue represents moderate severity. EPSS information is not available and the vulnerability is not listed in the CISA Known Exploited Vulnerabilities catalog. The likely attack vector requires an authenticated user on a shared deployment and allows that user to probe internal network services via arbitrary SMTP endpoints. The potential impact is internal reconnaissance that could facilitate further attacks on exposed internal resources. The overall risk is therefore moderate but should be addressed promptly.