The Intelligence Advanced Research Projects Activity issued a request for information on the possible threats and vulnerabilities associated with the use of large language models in intelligence operations.
In a notice posted Monday on SAM.gov, IARPA said the RFI will aid in the development of a framework to categorize the risks in LLM technologies.
The agency is looking to examine threats in “white box” LLM models, which means there is some privileged level of access to code or parameters, as well as “black box” models, which covers those with no privileged access to code.
IARPA is seeking feedback from organizations that may have novel methods to detect and mitigate LLM vulnerabilities, or approaches to quantifying their confidence in LLM results.
Interested parties may submit their response by Aug. 21.
