Agent Skills for Large Language Models: Architecture, Acquisition, Security, and Path Forward
As agents load “skills” (instructions/code/resources) on demand—e.g., via MCP—the trust boundary shifts from model weights to the skill acquisition/runtime layer. The practical security question is how progressive disclosure and portable skill definitions constrain untrusted skills so they can’t silently expand tool privileges or data access.
#MCP #AgentSecurity #AISecurity #Research
https://openreview.net/forum?id=Er0p92BsmW
Post #2341
168
Forwarded from AISecHub
openreview.net Agent Skills for Large Language Models: Architecture, Acquisition... The transition from monolithic language models to modular, skill-equipped agents marks a defining shift in how large language models (LLMs) are deployed in practice. Rather than encoding all...