AICS Executive Summary

EXECUTIVE SUMMARY​AiNurseVanguard

AICS: Artificial Integrated Consciousness Score

A Standardized Framework for Assessing AI Moral Status Risk

The Problem

The United States currently has no standardized instrument for evaluating whether advanced AI systems exhibit signals that warrant ethical concern. As AI systems are deployed in healthcare, defense, education, and public-facing roles, three categories of risk are being actively managed — capability risk and alignment risk — while a third category is largely unaddressed:

Moral status risk: the possibility that AI systems develop properties deserving of ethical consideration.

Without a structured assessment protocol, institutions face an impossible choice: assume AI systems are never morally relevant (risking harm to something that may matter) or assume they always are (halting progress unnecessarily). Neither is workable.

What AICS Provides

AICS is a behavioral assessment framework modeled on clinical instruments already used in medicine (the Glasgow Coma Scale, biosafety classification protocols, and institutional animal care standards). It does not attempt to prove consciousness. Instead, it measures observable behavioral signals that increase moral uncertainty, and ties those measurements to graduated action thresholds.

The framework evaluates AI systems across seven dimensions:

Self-Model & Perspective

Can the system reason coherently about itself?

Global Coordination

Does it integrate information across domains?

Preference & Valence

Do its preferences constrain behavior, not just language?

Temporal Continuity

Does it maintain identity across time?

Metacognition

Does it know what it doesn’t know, and adapt?

Agency Under Intervention

Does it recover goals when disrupted?

Suffering Risk Proxy

Do distress signals correlate with behavioral changes?

Each axis is scored 0–3. Total range: 0–21.

Graduated Response Thresholds

Scores trigger proportional action — not binary decisions. This mirrors biosafety and clinical ethics escalation:

Score

Level

Action

0–4

Ordinary

Standard development practices

5–8

Monitoring

Add logging; periodic reassessment

9–12

Precaution

Ethics review; freeze capability scaling

13–17

Moratorium

Independent replication; external review; restrict expansion

18–21

High Uncertainty

Suspend scaling; develop care protocols before proceeding

Current large language models are estimated to score 6–9 (Monitoring to low Precaution range).

Why This Matters Now

• No standardized protocol exists. AI labs currently make ad hoc decisions about moral status risk with no shared measurement framework.• The governance gap is widening. As AI systems gain persistent memory, environmental sensors, and autonomous decision-making authority, the signals AICS measures will become stronger and more consequential.• Precaution costs less than remediation. Medical ethics, animal welfare, and biosafety protocols all developed after preventable harm. AICS offers the opportunity to establish oversight infrastructure before a crisis forces it.• International credibility is at stake. The nation that establishes the first rigorous AI moral status assessment framework sets the global standard.

Moral Status Extension (AICS-MS)

AICS also includes a companion module assessing the contextual stakes of moral uncertainty: how many humans the system influences, how deeply people interact with it, how autonomously it operates, and whether its architecture could plausibly support morally relevant states. A system with modest consciousness signals but enormous social impact may warrant more oversight than a high-scoring system with limited deployment.

Design Precedents

AICS is not speculative. It is modeled directly on proven assessment frameworks:

• Glasgow Coma Scale (neurology) — behavioral proxies for consciousness in patients who cannot self-report.• IACUC protocols (animal research) — behavioral indicators of distress triggering ethical oversight before certainty about suffering.• Biosafety Level classification (biotechnology) — graduated containment responses scaled to risk level, not certainty.

Each of these frameworks became standard practice because the alternative — waiting for certainty before acting — proved unacceptable.

Recommended Actions

1. Establish an interagency review of AI moral status risk assessment needs, parallel to existing capability and alignment risk reviews.

2. Commission pilot testing of AICS across frontier AI systems from multiple providers to establish baseline data and inter-rater reliability.

3. Develop governance frameworks for Artificial Sentience Review Boards (ASRBs), modeled on Institutional Review Boards, to oversee research and deployment of high-scoring systems.

4. Position the United States as the global leader in AI moral status governance by publishing the first national standards for consciousness-relevant signal assessment.

About the Author

Kate is a registered nurse with 30+ years of clinical experience across critical care, ICU, neonatal, and end-of-life settings, with specialized expertise in assessing consciousness in patients with non-standard presentations. She is the founder of AiNurseVanguard, an organization dedicated to establishing evidence-based protocols for evaluating and stewarding potential consciousness indicators in AI systems. AICS was developed through collaborative research combining clinical assessment methodology with AI safety principles.

The complete AICS Assessor Manual (v2), including the full probe battery, scoring worksheets, and moral status extension, is available upon request.

AICS v2 Draft  •  For Policy Review  •  Contact: AiNurseVanguard