Content safety grounded in reasons, not just rules.

The questions we asked

  1. Warrant

    How can AI distinguish harmful content from satire, legitimate criticism, or cultural expression, and what evidence justifies its assessment?

  2. Legitimacy

    Where should the boundaries of content moderation lie, and how can they be justified without undermining freedom of expression or fundamental human rights?

  3. Legality

    How should AI moderation systems respond when national laws, cultural expectations, and international human-rights standards conflict?

  4. Practical context

    How can the same moderation system account for differences in language, culture, and application without applying inconsistent or arbitrary standards?

Rethinking the foundations of AI content safety

Content moderation is never merely a technical decision. What one person considers harmful, another may regard as satire, political criticism, or a legitimate expression of cultural identity. Yet moderation systems must also recognize genuine threats and protect people from abuse.

The Structural Modal Content Safety Framework investigates whether these competing demands can be addressed through a common normative structure. Drawing on structural normativism and our Responsible AI Framework, the project explores how fundamental ethical obligations can remain stable while their application takes account of legal, cultural, and situational differences.

Rather than relying exclusively on predefined categories and thresholds, the proposed architecture combines different dimensions of content assessment with explicit ethical justification and human review.

The underlying question is whether content safety can become more than a mechanism for deciding what to block: a process in which decisions can be understood, challenged, and justified.

Project status

In development