The S.O.R.F. Manifesto

Preamble

We, DevSimplex, recognize that artificial intelligence represents both unprecedented opportunity and significant responsibility. As builders of AI-powered systems, we hold ourselves accountable to the highest standards of safety, openness, responsibility, and fairness.

This mandate establishes binding principles for all AI systems developed, deployed, or operated under the DevSimplex name. It applies to all employees, contractors, partners, and affiliated entities.

The principles herein are not marketing statements. They are operational constraints that govern our engineering decisions, product development, and client engagements.

Article I: Safe

Section 1.1: Human Oversight

All AI systems capable of consequential decisions shall maintain human oversight mechanisms. No autonomous system shall operate in domains affecting human safety, employment, financial standing, or legal status without human-in-the-loop verification.

Consequential decisions include:

• Hiring, firing, or performance evaluation
• Credit, lending, or financial assessment
• Healthcare recommendations or triage
• Legal advice or case evaluation
• Educational grading or advancement
• Content moderation affecting user standing

Section 1.2: Fail-Safe Design

AI systems shall be designed with graceful degradation. When uncertainty exceeds defined thresholds, systems shall defer to human judgment rather than proceeding with potentially harmful actions.

Section 1.3: Security & Robustness

All AI systems shall undergo security assessment before deployment.

• Adversarial testing (prompt injection, jailbreaking)
• Data poisoning detection
• Model integrity verification
• Input validation and sanitization
• Output filtering for harmful content

Section 1.4: Incident Response

DevSimplex maintains documented procedures for AI incidents, including 24-hour internal notification, 72-hour client notification for significant incidents, and public disclosure for incidents affecting more than 1,000 users.

Article II: Open

Section 2.1: Transparency

Users shall be informed when interacting with AI systems. Marketing materials shall not misrepresent AI capabilities or limitations.

Section 2.2: Open Source Commitment

DevSimplex commits to releasing core infrastructure, SDKs, and non-competitive tooling as open source. We support the broader developer community through shared knowledge.

Section 2.3: BYOK (Bring Your Own Key)

Enterprise deployments shall support customer-managed API keys and on-premises inference options where feasible. Data sovereignty is a fundamental right.

Section 2.4: Model Transparency

When using third-party AI models, DevSimplex shall disclose which models are used, document model versions and updates, provide model switching options where feasible, and support open-source model alternatives.

Article III: Responsible

Section 3.1: Data Ethics

DevSimplex adheres to strict data ethics principles: minimal data collection, purpose limitation, clear retention policies, and user rights to access, correction, deletion, and data portability.

Section 3.2: Training Data Integrity

DevSimplex shall not train proprietary models on client data without explicit consent, use data marked as confidential for model improvement, or allow client data to “leak” into training sets of third-party models.

Section 3.3: Environmental Consideration

We commit to efficient inference, responsible training practices, and environmental consciousness in our infrastructure decisions.

Section 3.4: Prohibited Uses

DevSimplex shall not build, deploy, or knowingly enable:

• Autonomous weapons systems
• Mass surveillance without consent
• Social scoring systems
• Deceptive synthetic media without disclosure
• Systems designed to manipulate vulnerable populations
• Addiction-maximizing engagement systems

Article IV: Fair

Section 4.1: Bias Mitigation

AI systems shall be tested for demographic biases. Known biases shall be documented and mitigated. Systems with uncorrectable bias shall not be deployed for affected use cases.

Section 4.2: Accessibility

AI-powered products shall meet WCAG 2.1 AA accessibility standards.

Section 4.3: Economic Fairness

We commit to fair pricing, transparent pricing structures, educational and non-profit discounts, and clear upgrade/downgrade paths.

Section 4.4: Algorithmic Fairness

AI systems shall not discriminate based on protected characteristics including race, ethnicity, gender, religion, age, disability, or socioeconomic status.

Article V: Enforcement & Accountability

Section 5.1: Internal Review Board

An AI Ethics Review Board shall evaluate new AI products before release.

Section 5.2: Employee Responsibilities

All DevSimplex personnel shall complete AI ethics training annually, report potential violations, refuse to implement features violating this mandate, and document ethical concerns in technical decisions.

Section 5.3: External Audit

DevSimplex commits to annual third-party audits of compliance with this mandate.

Section 5.4: Client Accountability

DevSimplex reserves the right to terminate service for clients misusing AI systems, report illegal uses to appropriate authorities, and decline projects that violate this mandate.

Article VI: Amendments

Section 6.1: Amendment Process

This mandate may be amended by the DevSimplex Board of Directors with a 90-day public comment period for material changes.

Section 6.2: Version Control

All versions of this mandate shall be publicly archived on devsimplex.org, timestamped with effective dates, and accompanied by change summaries.

Section 6.3: Interpretation

Questions regarding interpretation of this mandate shall be directed to the AI Ethics Review Board. Published interpretations shall be binding.