Barrikade Core

Barrikade Core is a runtime security component for AI agents that combines prompt-injection detection with stateful session monitoring. Prompt-injection detection: - A staged pipeline normalizes obfuscation, checks signatures and embeddings, and applies machine-learning classifiers. - Uncertain inputs can be escalated to an optional local judge. - Each result is an allow, flag or block verdict with confidence, the deciding layer and optional diagnostics. - The API and Python SDK expose these primitives for integration. Stateful agent sessions: - A session starts with declared intent, permissions, provenance and an optional delegation chain. - Later inputs are compared with the declared task, and new domains and risky events deduct from a configurable risk budget. - Critical drift can pause a session and require escalation. - Supported intervention values are halt, downgrade, resample, escalate and revoke. - An in-memory store is included for local use; production persistence adapters remain deployment work. Evidence and telemetry: - Emits structured session events and can generate incident or near-miss reports covering provenance, permissions, tool activity, domains, drift assessments, risk deductions, interventions and outcomes. - Telemetry is available as structured JSONL with trace correlation and configurable sampling, intended as a foundation for integration with existing observability and SIEM workflows. Barrikade states that it is not a finished compliance product and does not present its internal benchmark figures as production guarantees.