Our Principles
The values and standards that guide every decision we make.
These twenty principles are the foundation of how Every Language Matters approaches both AI research and data annotation. They are not aspirational statements — they are operational commitments that shape how we hire, how we work, and what we refuse to compromise on.
Two disciplines. One shared commitment.
Our principles are split across the two core practices that define our work — AI research and data annotation. Together they form a complete picture of how we operate.
AI Research
10 principles for responsible AI development
Safety First
Prioritize the development of AI systems that are safe, reliable, and aligned with human values before optimizing for capability or speed to market.
Rigorous Empiricism
Ground all claims in evidence. Resist hype, maintain scientific humility, and publish negative results as readily as positive ones.
Long-Term Thinking
Make decisions with multi-decade consequences in mind. The impacts of foundational AI research compound over time.
Responsible Transparency
Be open about research methods, findings, and limitations while carefully managing dual-use risks that could enable harm.
Human-Centered Design
Build systems that augment and empower people rather than replace or diminish human agency, judgment, and dignity.
Inclusive Benefit
Actively work to ensure AI capabilities and their benefits are distributed broadly, not concentrated in the hands of a few.
Collaborative Ecosystem
Engage openly with academia, policymakers, civil society, and other labs. The challenges of AI are too large for any single organization to solve alone.
Ethical Accountability
Establish clear internal governance with real authority to pause, redirect, or halt work that poses unacceptable risks.
Respect for Human Oversight
Support the ability of humans — institutions, regulators, and society — to understand, monitor, and govern AI systems throughout their lifecycle.
Intellectual Integrity
Foster a culture where researchers can raise concerns freely, dissent is valued, and truth takes precedence over defending prior positions.
Data Annotation
10 principles for responsible data labelling
Data Quality as a Foundation
Treat annotation accuracy as non-negotiable. Flawed labels produce flawed models — quality at the data layer determines the ceiling of every system built on top of it.
Annotator Dignity & Fair Compensation
Recognize annotators as skilled contributors to AI development. Pay fairly, provide safe working conditions, and protect against exploitation.
Bias Awareness
Actively identify and mitigate cultural, demographic, and linguistic biases in labeling guidelines, annotator pools, and task design.
Clear & Consistent Guidelines
Invest in precise, well-documented annotation schemas. Ambiguity in instructions directly translates to noise in training data.
Diversity of Perspectives
Source annotators from varied backgrounds, cultures, and geographies to ensure labeled data reflects the full breadth of human experience.
Privacy by Default
Handle all data — especially personal, sensitive, or biometric data — with strict privacy protocols, minimizing exposure and ensuring regulatory compliance.
Feedback Loops with Researchers
Create direct channels between annotators and model researchers so that labeling challenges surface as signals, not silent errors.
Continuous Calibration
Regularly audit annotation consistency through inter-annotator agreement metrics, retraining, and updated guidelines as tasks evolve.
Ethical Task Screening
Refuse annotation work that involves harmful, exploitative, or deceptive content, regardless of commercial value.
Traceability & Auditability
Maintain full records of annotation decisions, versioning, and provenance so that data lineage can be audited, reproduced, and improved over time.
Research and annotation are inseparable.
These two sets of principles are not independent — they reinforce each other at every level of the work.
Data shapes models
The quality, diversity, and integrity of annotated data directly determines the safety and capability of any AI system. Research principles without data principles are incomplete.
People power both
Both disciplines depend on people — researchers who think rigorously and annotators who label honestly. Treating every contributor with dignity is a shared principle across both.
Feedback runs both ways
Research insights improve annotation guidelines. Annotation challenges surface research problems. The loop between the two is where the most important learning happens.
Community guidelines shape how we build AI systems.
These guidelines define how we conduct research, evaluate systems, and engage with communities. They evolve as the field advances and as we learn from real-world model behavior across languages and contexts.
Last updated — June 2026 · AI Research Governance · community@everylanguagematters.com