Evidence · DiggingBeagle record
SecureGARD: dynamic ontology-guided governance
Research on policy checks of AI-agent tool execution using semantic state and ontology assertions.
- Source role
- primary disclosure
Evidence record
Research on policy checks of AI-agent tool execution using semantic state and ontology assertions.
Claim-level citations (2)
- supportsSecureGARD inserted an external semantic policy check before agent tools ran: The SecureGARD authors report an observed 0% attack success rate across their selected benchmark tasks and LLM backbones; this does not establish protection against untested actions or an incomplete ontology.
Abstract; Main Results; Backbone Sensitivity
- supportsSecureGARD inserted an external semantic policy check before agent tools ran: SecureGARD proposes an ontology-guided external execution control that determines whether an agent’s requested tool action conforms to the current authorized semantic state.
Method: Semantic Execution Modeling; Execution Boundary and Semantic Adjudication