Agent security is composition security
NIST distinguishes risks created when model outputs are combined with software functions, data access and real-world actions.
FUURAA original conceptual visualWhat the evidence indicates
A concise reading of the source
NIST distinguishes risks created when model outputs are combined with software functions, data access and real-world actions.
FUURAA interpretation
Why this could matter
Security reviews must evaluate the complete agent loop rather than judging the language model in isolation.
How to read this signal
Documented development
The underlying event, report or finding has been published. Its future consequences may still be uncertain.



