Azure MCP exploited: Maliciously leaking user's KeyVault secrets to attackers

[sherdil2022]

Shouldn’t these ‘malicious’ prompts be part of testing? If we are allowing scripts to be executed through prompts, shouldn’t there be more checks and balances?

“Don’t trust user input” “Validate user input” “Sanitize user input” mantras ring in my head.

Why weren’t they ringing in those developers heads? Or were they vibe coding??

Tongue firmly in cheek - or may be not.

[emseetech]

Protecting against deterministic systems is hard enough, verbally defending against human language is an infinitely complex problem.

The only reasonable way to prevent this is to firewall your services.