The U.S. Marine Corps has released new guidance on the development, deployment and use of large language models and other generative artificial intelligence technologies within the service branch, Breaking Defense reported Friday.
Guidance for GenAI System Owners, Developers
Unveiled in December, the USMC Guidance on Generative Artificial Intelligence states that GenAI system owners and developers should obtain appropriate approvals for processing classified and sensitive information in compliance with existing software and container security policy and develop processes to document the source and attributes of training data before building or fine-tuning a GenAI model.
They should also test and evaluate GenAI systems in a controlled environment to ensure that such tools operate as expected and provide transparency and explainability for model outputs as required.
Instructions for GenAI System Users
GenAI system users should “distrust and verify all outputs prior to use” and use their best judgment when determining whether to add a citation.
The guidance holds users responsible for the data they input into publicly accessible GenAI tools. Users should also adhere to cybersecurity, legal, operational security, information and classification policies.
Establishment of AI Task Forces
The document directs commands to establish AI task forces or cells to help evaluate existing and in-development GenAI tools for applicability for use within the military branch.
Commands should also come up with a list of forthcoming preferred GenAI systems aligned with common use cases as a reference for USMC organizations seeking to meet their mission requirements with GenAI platforms.
According to the guidance, commands should ensure that developers, system owners and users use appropriate risk assessment frameworks for GenAI systems.
