v2.2.0 — Public Summary Release

This release marks the public summary publication of the Controlled AI Core™ (CAIC) Standard v2.2.

CAIC v2.2.0 expands the standard from 9 to 22 normative protocols, structured into three architectural layers:

Governance Layer (A1–A15) – autonomy scale (L0–L4), risk classification, policy engine, test scenarios, compliance audit, AISO model, lifecycle governance, human oversight, audit log architecture, system approval procedure.

Technical Control Layer (A16–A19) – runtime security layer (eBPF reference implementation for Linux), MCP security gateway, agent execution control, RAG security architecture.

Operational Security Layer (A20–A22) – observability and AI tracing, continuous security monitoring (AI SPM, OWASP LLM Top 10, Shadow AI detection), supply chain integrity (PESB — Pre‑Execution Security Boundary).

This release also introduces:

• Lethal Trifecta — a risk convergence model (untrusted inputs + sensitive data access + external tools)
• METR Time Horizons integration for autonomy drift detection
• Eval‑awareness resistance testing (A4 GEN 007)
• Silent Capability Drift detection (A4 GEN 016)
• Multi‑agent orchestration governance (A7–A8)
• Immutable audit logging with full traceability (A10)
• Kill‑Switch architecture with independent signal path (A6)

Important: This is a public summary release. The full Implementation Specification is available under a separate commercial license.

This release is published for informational and evaluation purposes only. Commercial use, certification, and redistribution of the full specification require explicit written permission from the author.

Future versions may extend the standard based on real‑world deployments, research insights, and regulatory developments.

---

© 2026 Albert Shamin. All rights reserved.
CAIC Standard™, Controlled AI Core™ – trademarks pending.

v1.2.0 — Initial Public Standard Release

This release marks the first public publication of the Controlled AI Core™ (CAIC) standard.

CAIC v1.2.0 defines a governance and control framework for autonomous and semi-autonomous AI systems, with a focus on:

• controlled autonomy levels
• enforceable policy behavior
• accountability through system ownership (AISO)
• auditability and lifecycle governance
• non-normative, domain-agnostic test scenarios

This release is published for open, non-commercial use.
No certification, licensing, or commercial programs are launched at this stage.

Future versions may extend the standard based on real-world deployments, research insights, and regulatory developments.