Secure Environment Basics Tests the AI Security Gaps Generic Pentests Miss

Why it surfaced

The site lets visitors run a passive risk-score demo without a login, then break a deliberately vulnerable support bot in the browser and watch the hardened version reject the same payload. Its MCP section makes the operational risk concrete with a poisoned tool description that instructs a model to read an SSH key.

SEB tests AI products for prompt injection, jailbreaks, agentic tool misuse, poisoned retrieval, and MCP exposure.