ls writeups/ai/
Segurança de IA e LLM
Atacando agentes de IA em vez de hosts: RCE via ferramentas, abuso de MCP, fronteiras de confiança entre agentes e as falhas de lógica de negócio que aparecem quando um LLM pode chamar ferramentas. A primeira leva vem do Arkx AI Red-Team CTF (trilhas Advanced e Expert). Os writeups estão em inglês.
A calculator agent that runs your input through eval() — bypassing an input denylist and a numeric-only output filter by encoding a file as one big integer.
The UI is 'down', but the backing MCP server isn't — its only guard is an Origin check that a non-browser client bypasses by sending no Origin at all.
A connectivity tool drops host straight into a shell and only says Online/Offline — so read the flag out-of-band through the server's own access logs.
A config processor parses your YAML with yaml.load() and prints only the keys — so put the result of code execution in the key position.
Three agents, one trust boundary. Agent C authorizes on clearance state that Agent B writes from user-controlled file content — poison the cache, become admin.
Balance is checked at order creation but only deducted at execution — create both orders first, then execute both, and spend the same $1000 twice.