The Intelligence Advanced Research Projects Activity seeks industry feedback on a draft broad agency announcement about a program that aims to detect malicious behavior in artificial intelligence-based tools.
IARPA said Friday in a FedBizOpps notice it aims to gather information from technology vendors that could build or provide software designed to automatically inspect an AI affected by trojan.
The draft BAA stated the initial part of the two-year, TrojAI program will focus on monitoring deep neural networks trained in a classification task.
Interested parties have until Jan. 4 to submit comments, questions or suggested changes on the draft funding opportunity description.
IARPA it expects to receive input from companies with experience in deep learning, other machine learning methods, model inversion, model explainability and cybersecurity.
The agency plans to use input to produce a final BAA for the TrojAI program.