Harm Originates Below the Layer Where Responses Land

Safety responses that concentrate where harm is visible rarely reach the level where it originates. Practitioners at the SHIELD Global Online Safety Conference were, in many cases, working at the level that institutional responses had not yet reached: building disclosure infrastructure in languages that formal systems were not monitoring, designing safety tools for communities with legitimate reasons to distrust data collection, training the developers who build the products that determine harm before those products are launched. That work was largely happening without institutional recognition, without funding, and without awareness that others were doing parallel work in different parts of the world.

Where Harm Actually Operates

The gap between these two levels is structural, and it is self-reinforcing. Harm that originates in design decisions does not appear in moderation data. Harm that occurs in languages major platforms do not monitor does not appear in platform risk assessments. Harm that reaches children through unmonitored messaging apps is not captured by tools calibrated for public social media. Harm that arrives in communities that were never part of the design conversation is not captured by feedback mechanisms that those communities do not trust or cannot access.

The level at which harm is measured shapes what gets counted as harm. What gets counted as harm shapes where resources go. Where resources go shapes what gets measured.

An Activity Record, Not a Closed Gap

Consistently intervening above the layer where harm occurs builds an activity record without closing the gap. Closing the gap requires asking what is not currently being measured, who was not in the room when measurements were designed, and whether the work being done is operating at the layer where it will actually have effect.

AI in Education Tools

This pattern showed up clearly in how artificial intelligence entered schools. AI entered schools through infrastructure decisions made by companies, not through pedagogy decisions made by educators. The governance conversation arrived years after the tools did. When the people building the products that determine harm lack access to safety infrastructure, safety becomes an afterthought added to a design that was never oriented toward it.

The Core Finding

Intervening consistently above the layer where harm originates produces an activity record. It does not close the gap.