Last updated: November 27, 2025

Ethical AI Policy

We believe AI must be developed and deployed responsibly. Our mission to protect children online carries deep ethical responsibilities. This policy outlines our commitments to ethical AI development, deployment, and governance.

Child safety firstPrivacy by designHuman oversight

Our Commitment

AI should protect children without compromising privacy, fairness, or transparency. We pair on-device design with rigorous governance to achieve that.

These commitments cover our website, SDKs, APIs, documentation, and related services using RoseShield technology.

Core Ethical Principles

Guiding values

Child Safety First

Our primary mission is protecting children from online harm. Every decision is evaluated against this purpose.

Design systems that prioritize child welfare above commercial interests
Continuously improve detection accuracy
Respond promptly to emerging threats
Never compromise safety features for convenience

Privacy by Design

Child protection and privacy are not mutually exclusive. Our architecture ensures both.

Process all data on-device—no cloud transmission required
Collect zero personal information from end users
Reject surveillance-based approaches
Protect privacy as a fundamental right

Fairness & Non-Discrimination

AI must work equitably across backgrounds, languages, and demographics.

Test and mitigate algorithmic bias continuously
Ensure protection works across languages and cultures
Avoid profiling based on protected characteristics
Provide equal protection to all children

Transparency

We are open about how our technology works, its capabilities, and its limitations.

Document how our AI makes decisions
Acknowledge limitations honestly
Provide clear information about what we detect and how
Never overstate capabilities or make false claims

Human Oversight

AI should augment human judgment, not replace it.

Enable developer configuration and control
Support human review of flagged content when appropriate
Provide clear escalation paths for edge cases
Never make irreversible decisions without human oversight

Minimizing Harm

We actively work to prevent unintended harm.

Balance protection with avoiding over-blocking
Consider psychological impact of warnings
Avoid false accusations or unjust penalties
Continuously evaluate for unintended consequences

What We Will Never Do

Non-negotiables

These are hard lines we will not cross, regardless of commercial pressure or technical capability:

Mass surveillance or tracking of children's online behavior
Collecting biometric data (facial recognition, voice prints)
Building profiles or dossiers on individual users
Sharing or selling user data to third parties
Using AI to manipulate or exploit users
Discriminatory targeting based on protected characteristics
Deploying unvetted or experimental AI in production
Prioritizing commercial interests over child safety

Ethical Governance

Oversight

Regular Audits

Our AI systems undergo regular internal and external audits for bias, accuracy, and ethical compliance.

Incident Response

We maintain clear procedures for addressing ethical concerns, mistakes, or unintended harms.

Stakeholder Input

We seek input from child safety experts, privacy advocates, and affected communities.

Continuous Learning

Our ethical guidelines evolve as we learn more about AI impacts and best practices.

Honesty About Limitations

Be transparent

No AI system is perfect. We are transparent about the limitations of our technology:

AI detection may occasionally produce false positives or negatives
New threat patterns may not be immediately recognized
Context and nuance can be challenging for automated systems
Our technology supplements but does not replace human judgment
Regular updates are needed to address evolving threats

Reporting Ethical Concerns

Escalation

We encourage anyone to report ethical concerns about our AI systems. Reports can be made anonymously and will be investigated promptly. We are committed to addressing legitimate concerns and making necessary improvements.