AI Security Glossary
A working reference for agent security, governed execution, and cryptographic verification concepts.
Threat Vectors
Adversarial surfaces at the intersection of LLM reasoning and the Model Context Protocol.
Prompt Injection
Foundational LLM vulnerability involving instruction override.
Indirect Injection
Adversarial instructions embedded in retrieved external content.
SSRF (AI Context)
Agent-driven exploitation of cloud metadata and internal services.
Tool-Call Hijacking
Unauthorized invocation of sensitive MCP tool capabilities.
Data Exfiltration
Covert transmission of context data via outbound tool parameters.
Context Poisoning
Persistence-oriented attacks against agent memory and beliefs.
LLM Jailbreak
Bypassing model-layer behavioral constraints via engineering.
Path Traversal
Unauthorized filesystem access via tool parameter manipulation.
Privilege Escalation
Acquisition of unauthorized capabilities via semantic drift.
Permission Drift
Incremental expansion of agent scope beyond initial intent.
Architecture & Primitives
Cryptographic and architectural constructs that form the ProvnAI governance layer.
Governed Execution
Policy-checked, evidence-producing control for real agent actions.
Evidence Capsule
Cryptographic evidence construct preserving the governance context of governed execution.
Deterministic Proxy
Binary, rule-based enforcement at the proxy boundary.
Merkle Audit Trail
Tamper-evident, append-only cryptographic logging.
Execution Boundary
Formally defined perimeter for authorized agent actions.
Authority Control
Binding actions to verified principals and intent scopes.
Intent Attestation
Binding of agentic sessions to authorized goals.
TEE Isolation
Hardware-enforced enclave isolation for agent runtimes.
MCP Tool Manifest
Signed registry of authorized tool schemas and capabilities.
Per-Action Verification
Per-action verification architecture for autonomous AI.
Witness Log
Tamper-evident execution record preserving what crossed the boundary.
MCP Standard
The open standard for AI model-resource interaction.
VEX Protocol
Governance layer for real AI actions.