Operational Ethics

Operational Ethics — Cognition Strategy Group

Document Class: Binding Operational Instrument
Revision Cycle: Annual Mandatory Review
Governing Authority: Cognition Strategy Group Ethics Board

1.1 Foundational Ethical Posture

Cognition Strategy Group (hereinafter "the Firm") operates under the position that the deployment of large language models, autonomous pipeline systems, and AI-augmented decision infrastructure at enterprise scale carries non-delegable ethical obligations. These obligations are not aspirational. They are operationally enforced through architectural constraint, contractual obligation, and ongoing audit.

This document constitutes the Firm's binding statement of ethical operating principles. Deviation from these principles by any Firm personnel, Sub-Processor, or automated system operating under Firm authority constitutes a material breach subject to formal remediation and, where applicable, Mandate termination.

1.2 Prohibited Deployment Categories

The Firm will not accept, initiate, or continue Mandate Engagements involving the deployment of AI systems for the following purposes, without exception:

• Mass surveillance of individuals without explicit informed consent and lawful authority

• Autonomous lethal decision-making or any system whose outputs directly govern the application of physical force

• Manipulative inference targeting — systems designed to exploit psychological vulnerabilities to compel behaviour contrary to the subject's rational self-interest

• Discriminatory allocation systems that produce demonstrably disparate impact across protected characteristics without lawful justification and rigorous bias mitigation

• Synthetic media generation designed to deceive without disclosure, including deepfake systems deployed without explicit consent frameworks

• Unconstrained autonomous agents operating in production environments without human oversight mechanisms and interpretable circuit-breaker controls

1.3 Bias Auditing and Fairness Obligations

All classification, allocation, and recommendation systems deployed under Mandate must undergo a documented bias audit prior to production deployment. The audit must assess:

• Demographic parity across all protected characteristics relevant to the deployment context

• Equalised odds for high-stakes decision outputs (credit, employment, healthcare, legal)

• Calibration integrity — ensuring model confidence scores reflect empirical accuracy across subgroups

• Counterfactual fairness — outputs must not materially differ for otherwise identical inputs differing only in protected characteristic values

Bias audit findings are disclosed to the Client in full. Deployments with unresolved high-severity bias findings do not proceed to production without explicit Client acknowledgement of risk and documented mitigation commitment.

1.4 Human Oversight Requirements

The Firm mandates that all production AI systems deployed under Mandate operate with documented human oversight mechanisms. These include:

• A designated Model Governance Owner on the Client side with authority to suspend automated decision pipelines

• Confidence-gated escalation — automated outputs below defined confidence thresholds are routed to human review

• Circuit-breaker controls — configurable thresholds that trigger automatic pipeline suspension upon anomaly detection

• Mandatory review cadence — no production system operates more than ninety (90) days without a formal human performance review

1.5 Transparency and Explainability Standards

The Firm commits to deploying systems whose decision logic is interpretable to the degree technically feasible given model architecture. For high-stakes deployment contexts, the Firm requires:

• SHAP or equivalent attribution methods for classification systems affecting individuals

• Natural language explanations generated alongside automated decisions subject to regulatory challenge

• Audit trail completeness — every automated decision is logged with sufficient metadata to reconstruct the inference pathway

1.6 Ethics Incident Reporting

Any Firm personnel, Client personnel, or third party who identifies an ethical violation, near-miss, or systemic risk within a Mandate deployment is obligated to report via the Firm's Ethics Incident Register. Reports are investigated within five (5) business days. Substantiated incidents are disclosed to the Client in writing and, where legally required, to relevant regulatory authorities.