What happened
NeuralTrust launched (Aug 5, 2026, Black Hat) a runtime security mesh built on its Agent Gateway that inspects agentic traffic across six stages — prompt intake, tool calls, tool/MCP responses, tool authorization, reasoning/tool-selection drift, and final output — without requiring custom per-agent integrations.
Why it matters
Addresses a real gap: most AI security tools check declared agent behavior (system prompts/policies) rather than actual runtime actions across the full agentic loop, including MCP tool-call inspection and reasoning-drift detection.
Applicability
Security teams deploying heterogeneous AI agent fleets (multiple frameworks/clouds) should evaluate this as a cross-cutting runtime enforcement layer vs. per-tool point solutions.