Seguridad de IA
Practical security testing for applications that use AI or language models, focused on prompts, exposed information, restrictions, and connected functions.
We test the AI feature as a user would, attempting to bypass its instructions, obtain information that should not be shown, or trigger unintended results.
The scope is defined around the functions available in the application. It does not include reviewing the model, training data, or internal source code.
Prompts that attempt to ignore rules or change the intended behavior
Responses that may reveal sensitive, private, or internal information
Basic checks for outputs the application is expected to restrict
Checks that available actions cannot be used outside their intended purpose