The Intelligence Advanced Research Projects Activity will hold a proposers day on Oct. 24 in Washington, D.C. to discuss with the industry a new program that aims to identify the safe uses of large language models for applications relevant to the intelligence community.
A notice posted Friday on SAM.gov states that the BENGAL initiative seeks to develop novel technologies and methodologies to detect, characterize and mitigate threats and vulnerabilities associated with using LLMs.
Program participants will create a taxonomy of LLM threat modes within their selected topic domains and develop tools to probe LLM models to identify biases, threats or vulnerabilities.
The U.S. government plans to use LLMs for a range of applications including the rapid summarization and contextualization of IC documents.
IARPA will introduce topic areas and additional program requirements on the proposers day.
Interested parties have until Oct. 18 to register for the event.
